memes, networks and artifacts. cultural circulation in an artificial peer-production society

Ynonamous Committee

Memes, networks and artifacts

Cultural circulation in an artificial peer-production society

February 16, 2011

Springer

Perché dovremmo studiare questo problema se non cidivertiamo a farlo?

(Nicola Cabibbo)

Contents

1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

2 Peer production, Web2.0: networked economy and culture . 52.1 A new information economy? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52.2 Accessing and making sense of information . . . . . . . . . . . . . . . . . 9

2.2.1 Flickr Commons: knowledge goes online, live. . . . . . . . . . 112.2.2 Community judgment, common law and democracy 2.0 13

2.3 “Don’t worry, we’re from the Internets” The persistence ofculture in online communities . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

3 The Meme-To-Web model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233.1 Agent based simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243.2 Related works: models of blogging behavior . . . . . . . . . . . . . . . . . 253.3 Meme-to-Web. An abstract, generative, agent-based model of

peer-based content production and access . . . . . . . . . . . . . . . . . . 273.3.1 Model internals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 283.3.2 Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 323.3.3 Model falsification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343.3.4 Strengths and limits of the model . . . . . . . . . . . . . . . . . . . 343.3.5 What to look for and why . . . . . . . . . . . . . . . . . . . . . . . . . . 35

4 Build a network of artifacts, (random) walk on it . . . . . . . . . . 374.1 Publishing phase: network growth and similarities . . . . . . . . . . . 384.2 Memes and minds: looking for infodiversity . . . . . . . . . . . . . . . . . 39

5 PageRank and its discontents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 415.1 Googlecracy, or the “search engine bias” . . . . . . . . . . . . . . . . . . . . 435.2 Traffic distribution in the artificial network with PageRank . . . 455.3 Effects of PageRank on the production phase: network shape. . 485.4 Memes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

VIII Contents

6 Collaborative filters. The wisdom of crowds . . . . . . . . . . . . . . . . 51

7 Discussion and (hopefully) future work . . . . . . . . . . . . . . . . . . . . 57

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

1

Introduction

These lists [generated by google] can be dangerous – not for oldpeople like me, who have acquired their knowledge in another way,but for young people, for whom Google is a tragedy. Schools oughtto teach the high art of how to be discriminating. Teachers shouldinstruct students on the difference between good and bad

Umberto Eco1

It turns out that we are not intellectual lemmings.

Yochai Benkler

In the last two decades the Internet has become, at the same time, (a) themain repository of the symbolic production of (at least) western societies and(b) a platform for decentralized, collaborative and usually non market-based,content production - a phenomenon sometimes referred to as ’networked in-formation economy’, or ’peer production’. With regards to (a), there is littleor no doubt that almost the entire cultural production of mankind can be ac-cessed via the Internet. The question from which this work moves is, sic andsimpliciter, how the access takes place. It is by no means a trivial one: thepractices and architectures that persede over the access to knowledge becomecharacterizing traits of the societies of reference at various levels. Think ofthe written word restructuring the social ladder of the Fertile Crescent, withthe introduction of a new class of professionals. Another example is copyright,a normative framework produced by the diffusion of the printing press thatstill constitutes an annoyance nowadays, when the printing press no longer isthe main mean of information transmission. So what are the internet-basedpractices and architectures that implicitly mediate our access to knowledge?What their long-term effect on the cultural evolution of society? The claim wemake in this work is that the Internet - and the World Wide Web especially -

1 Spiegel interview with Umberto Eco, retrieved from http://www.spiegel.de/international/zeitgeist/0,1518,659577,00.html

2 1 Introduction

as a medium for storage and transmission of knowledge fundamentally relieson a set of peculiar architectures that exploit the unattended inter-action of large amounts of faceless, dispersed, heterogeneous users(the crowds) to provide meaningful categorizations of content withrelation to the whole corpus of production. These platforms (Google’sPageRank, the numerous collective content filters, social bookmarking sites,micro-broadcasting applications, collectively referred to as wisdom of crowdapplications) are tools emerged in the first place to provide a shortcut toinformation in the overloaded peer-production web. They respond mainly toengineering constraints, yet play a crucial epistemic function insofar providinga distributed, emerging classification and organization of knowledge, reshap-ing the way we approach the symbolic production of our society, replacingthe role of respected institutional authorities, and their complementary side,which is shared social trust. This may be argued to interfere with dimensionsof some importance in the history of western civilization, like the division ofintellectual labor, the entrustment on professional expertise and on the insti-tutions of culture, education and science. No surprise that it bewilders highprofile polymaths nurtured in such environment. The widespread, fundamen-tal role of these platforms calls for deep, multidisciplinary investigation of thedifferent aspects of the phenomenon. If it’s trough these platforms that weare to organize the circulation and organization of our ideas, we might wellbe interested in knowing whether the architecture and practices that theseplatforms employ encourage conformity, or favor innovation, as an example.What is the fate of critical opinions, of dangerous ideas, of fringe politics in aworld where Wisdom of Crowds algorithms select the good and the bad? Um-berto Eco has no doubts, but different voices can be heard. “A Culture BothPlastic and Critical ”, one which allows for several interpretations of a sameobject to coexist and equal opportunities for most of such interpretations tospread, is the outcome of the mediation of collective architectures – accordingto philosopher Yochai Benkler.

How this work is structured

This work seeks to explore these questions and test the validity of some ofthe claims above, tentatively applying a fascinating analytical method for thesocial sciences: one that aims at reproducing the investigated phenomenoninstead of simply observing it. We will build a society of artificial cognitiveagents endowed with a certain cultural configuration. A subset of the agentswill produce cultural artifacts that will carry some traits of the cultural mi-lieu of the creator, in an abstract implementation with each artifact being nomore than a set of memes - basic items of knowledge existing in the form ofatomic signs contained in agent’s minds and embedded in cultural artifacts.Memes - simultaneously contained in the minds of the agents and in the arti-facts that they produce and link - represent the ideas and concepts circulating

1 Introduction 3

in the society. As the system runs, agents create artifacts and connect themwith other artifacts, creating a network of directed links that simulates thefundamental structure of the world wide web. Networked artifacts, just likewebsites, will be accessible to agents through the mediation of certain mech-anisms representative of those that organize our knowledge and our access toit on the Web, with a focus on the most influential search engine algorithm,Google’s PageRank, and a prototypical social content website, modeled afterReddit/Digg. Agents exploring the network of artifacts, will “read” some ofthem and, under certain circumstances, internalize some of the memes con-tained. The analysis will proceed on three levels: we will (a) examine thestructure and the features of the network of artifacts emerging from each dif-ferent information filtering platforms, comparing them to those empiricallyobserved in publicly available web datasets; (b) explore the distribution ofusers’ attention among artifacts that each filtering application triggers andconfront our outcomes to traffic dynamics as described in existing literature;(c) examine the distribution of memes that the different platforms produce inthe population, making assumptions on the rate of infodiversity - a conceptthat we introduce, in analogy with the biological concept of biodiversity, as ameasure of the memetic variability found in a population - that each platformdetermines.

In the next chapter we describe the emergence of a new paradigm in cul-tural production triggered by the mass diffusion of the personal computerand broadband connections: that of peer-based cultural production. We willdig into the phenomenon in an economic perspective, through the work ofYochai Benkler, and in a cultural historic perspective, reviewing a few recentinsightful works.

In Chapter 3 we introduce our method of analysis and make some hyper-bolic assumptions on the individuals’ cognitive and behavioral habits relatedto the exploration and the fruition of online content. We propose a modelof individual behavior drawing on existing literature and on our own naiveassumptions and build a computer simulation2 of it.

In the subsequent chapters we let the simulation run, progressively en-riching it with the developments of information filtering technologies in theorder of historical appearance. Every technology will be described in its mainfeatures, relevant recent research will be reviewed and an abstract implemen-tation will be proposed for the scope of the work. The technologies will betested for the degree of infodiversity they appear to favor and the networksthey give shape to. We aim to generate - in scale - the evolution of wisdomof crowds platforms starting from their origins, in order to gather a bird-viewsight over the path taken by ’the Internet’ in terms of cultural plasticity andvariability. The goal, as we will see, is still far to be obtained, but with the ef-fort presented here we aim to propose a method and paint the road for a seriesof investigations on an issue whose cruciality appears to be underestimated.

2 Source code in Appendix and online at http://labss.istc.cnr.it/~stefano

2

Peer production, Web2.0: networked economyand culture

2.1 A new information economy?

Fig. 2.1. Time Magazine “person of the year” edition. Years 1983 and 2007

The two covers of Time Magazine reproduced above mark the fundamen-tal turning points of the socio-economic process discussed in this work: themass diffusion of a non specialized device for the production of cultural arti-facts, anticipated in the 1983 edition, and the ultimate affirmation of a newparadigm in content production, resulting from said devices being networkedtogether, affirming itself 25 years later.

6 2 Peer production, Web2.0: networked economy and culture

The personal computer began its journey into American households andlives in the early ’80s, to become almost ubiquitous at the end of the ’90s. Bythe end of the past decade the world was ready for yet another media revolu-tion, springing from the pairwise diffusion of home computers and broadbandnetwork connections: that of user generated content. In these twenty-five years,as Tobias Schäfer resumes in his PhD dissertation on “Bastard Culture”,

the machine initially developed for solving complex and repeti-tive arithmetic problems developed into a common office device, andsubsequently into an everyday medium for consumers who can dopractically anything with it, what can be informational formulated,from filing tax returns to organizing holiday pictures. The Internetand its successful application, the World Wide Web (WWW), havebeen crucial in this development. The WWW has enabled large mediaaudiences to recognize the computer as a handy tool for communica-tion, entertainment and leisure activities. Software like web browsers,which embed networking in a graphical user interface, attractive ser-vices such as web-based e-mail, chat programs, online communities,and Internet forums have increased the computer’s appeal to a largegroup of consumers. The Internet diffused aspects of the computer sothat not only machines but also people became globally connected.The networked computer has become a commonly used medium inthe Western industrialized countries. [50]

The World Wide Web, one of the technologies that sit on top of Internet,grew exponentially thanks to its ease of use and greater commodification op-portunities1, first aggregating functionalities once dispersed among multipleplatforms2 and slowly becoming home to personal publishing platforms, suchas weblogs, and later to large video and image hosting sites3. The 2007 edition

1 It is our persuasion that the hierarchical nature of the Web - extremely hierar-chical if compared to other networks operating on top of the Internet, say peerto peer ones, like BitTorrent - renders it more suitable for commercial exploita-tion, because a pure client-server architecture gives the owner of the server themaximum control over users’ activity and data. The tendency towards centraliza-tion in a once completely decentralized environment (consider that much of thediscourse on the intrinsic freedom and horizontality of the Internet is due to itsdecentralized architecture) is clearly visible if one looks at Facebook. Facebook isa walled garden where many of the activities once performed via different appli-cations and independent networks (chat, email, even e-commerce) are enclosed ina single architecture using a star network topology, with ideally one single centralserver.

2 Think of web based email gradually replacing traditional IMAP and POP basedemail, web boards replacing Usenet, later the “Web as a platform” hype, with webbased office applications such as word processors trying to bring online activities(and their related data) traditionally performed offline on individual computers.

3 http://youtube.com was founded in 2005, so did Google Video

2.1 A new information economy? 7

of the person of the year celebrates a collective You - the new individual con-tent creator - that challenges the traditional broadcast principle of centralizedcontent production and distribution. It was the celebration of the so calledReadWriteWeb, or Web2.0 4 encompassing weblogs, social bookmarking, wikisand other technologies, seen as marking a distinctive shift from earlier, sup-posedly less participatory, web technologies. We won’t go into detail discussingthe rhetoric of participation and the potential deception that it carries. Weagree with the conclusions of Schäfer himself: a truly participatory culture hasyet to emerge as a complex dispositif, not limited to users participating byappropriating commercial media texts or publishing their own productions.

What is of interest here though is the double effect5 triggered by the easypublishing and content sharing that Web2.0 applications bring around: the un-precedented amount of content produced by regular, mostly non professional,users and distributed via social platforms and the acceleration of an alreadyongoing phenomenon: the progressive transportation on the Internet of al-most the entire cultural production of mankind. Together the cheap personalcomputer and the free (of charge) and accessible publishing platforms are thetwo pillars of a yet to establish, but surely emerging, economic and culturalparadigm, that of the Networked information Economy / Peer-production So-ciety. The definition is of Yochai Benkler, whose enthusiastic point of viewon the political and economic implications of the changing mediascape is wellworth a review.

Fig. 2.2. Yochai Benkler

Yochai Benkler has been called "the lead-ing intellectual of the information age" [49].He proposes that volunteer-based projects,flourishing outside or in overt denial ofmarket-based dynamics, such as Wikipediaand GNU/Linux, are the next stage of hu-man organization and economic production.In his Wealth of Networks [8] Benkler movesfrom the notions of static/dynamic efficiencyand transaction costs to outline the profoundchanges that de-materialization has broughtto the traditional industrial-based economicinterpretation of production. Computers, forone, are surely the main means of produc-tion of the immaterial goods central to theeconomy of this century. Now in Benkler’sview, their universal diffusion equates to aform of distributed ownership of the means

4 The term Web 2.0 was coined by publisher Tom O’Reilly to describe Internettechnologies, summarized as Asynchronous Java and XML (AJAX) and the newservices evolving around the use of it.

5 Effect or cause? Causality is tricky here


of production, a situation with little or no antecedents in recent history: notprivately held and concentrated in the hands of few, as in a standard capi-talist economy; not collectively held, as it could be in a state-run economy.Instead the capillar diffusion of a highly granular machinery as the personalcomputer removes the barriers to entrance that have protected the culturalindustry for decades, in the form of concentration of means of production. Ifwe keep in mind the traditional economic interpretation of production, theCobb-Douglas function:

Y = ALαKβ

where Y=output, L=labor input, K=capital input, in an immaterial, knowl-edge based economy, not only the means of production, (capital) are in theunique position of being almost universally diffused and accessible, but alsothe other component of the K, raw matter, happens to be in a very appeal-ing condition: that of being a common. The production of cultural goods, infact, relies heavily on pre-existing knowledge (“shoulders of giants effect”, inBenkler’s words), in the form of works of art, books, music, thought, thatmake “derivative works” possible. The ownership of such first matter is *the*real challenging point of this century, argues Benkler. In the previous modeof production such “raw matter” had progressively concentrated in the handsof the cultural industry, because of the huge investments needed to transformit into the final product and to organize its mass distribution. The “digitalrevolution” subverts this situation: not only every single piece of informationis potentially accessible at no cost, everybody can transform at very little costthe old information - using cheap computers and finding a limit only in onesown capabilities - but can also distribute a derivative work, still at virtually nocost, via the internet and its free platforms. Here comes the aversion for copy-right extensions, strict intellectual property and patents on ideas, methodsand software. These institutes, far from encouraging creativity and stimulat-ing the production of new content, are of huge obstacle, in that they aimat artificially creating scarcity in a situation of abundance. Benkler equatescopyright extensions to the enclosures movement of the early XVII century, inthat they let a class of individuals appropriate common goods, preventing thecommunity from usefully employing common resources for their own benefit,and in this case, creativity.

However the emerging new organization of economy, withWe can sketch a model of the peer based production economy and its main

differences with the cultural industry model (Table 2.1)The model we present in Chapter 3 strongly draws on the utopical society

prefigured by Benkler, the ’peer-production’ condition. We designed a societywith a substantial fraction of content producers, a constant inflow of newagents and high porosity between the group of producers and non producers.Object of our investigation will be the item number four, the content filteringapplications.

Let us now briefly discuss them and their relevance.

2.2 Accessing and making sense of information 9

Table 2.1. Broadcast and peer production

# BROADCAST PEERPRODUCTION

1 Sources Few Many

2 Barriers to entrance High Low

3 Producers / consumers separation Nect Fuzzy

4 Content selection and distribution Centralized, prior topublication

Decentralized,emerging,

postpublication

2.2 Accessing and making sense of information

Philosopher Gloria Origgi, in a thought provoking talk at the College deFrance [43], advanced a somewhat uncommon interpretation of the role of In-ternet in storing and accessing knowledge. Not only a communication medium(or set of media), according to the Italian scholar, the Internet has to be inves-tigated first of all as the set of technologies that we employ to store, organizeand access the knowledge produced by the society. Origgi calls this set of tools“meta memory” and, in line with the tradition of cultural anthropology, putsthe Internet at the end of a journey that featured the spoken word, writing, theprinting press, and finally other electronic media - means of communicationand information storage that the internet integrates and hybridates.

Writing [...] is an external memory device that makes possible thereorganization of intellectual life and the structuring of thoughts, nei-ther of which are possible in oral cultures. With the introduction ofwriting, one part of our cognition “leaves” the brain to be distributedamong external supports.Printing [...] redistributes cultural memory, changing the configurationof the “informational pyramid” in the diffusion of knowledge.

The Web shares similarities with both writing and printing, being an externalmemory device that distributes the cultural memory in a population, but

unlike writing and printing, the Web presents a radical change inthe conditions for accessing and recovering cultural memory with theintroduction of new devices for managing meta-memory, i.e., the pro-cesses for accessing and recovering memory

In a perspective of history of culture, the Internet is a peculiar instrument(medium) for a number of reasons. Apart from being characterized by an un-precedented storage capacity, the Internet replaces different media and hostsevery single possible format. Audio, video, writings are stored and diffused inthe exact same fashion. The reticular organization of information inspired all


sorts of fascinating metaphors, like that of "distributed collective intelligence":an extended brain wrapping the whole planet, with every connected individ-ual as a braincell synaptically connected to others by the network cables [35].Anyway the most breakthrough trait of the internet as an information storagefacility are the unprecedented approaches to content selection, managementand retrieval, i.e. the meta-memory that selects and ranks every single itemof information with relation to the whole corpus of production. In Origgi’swords:

Culture, to a large extent, consists in the conception, organization andinstitutionalization of an efficient meta-memory, i.e. a system of rules,practices and representations that allow us to usefully orient ourselvesin the collective memory. A good part of our scholastic educationconsists in internalizing systems of meta-memory, classifications ofstyle, rankings, etc.. chosen by our particular culture.[...]Meta-memory doesn’t serve only a cognitive function – to retrieveinformation from a corpus – but a social and epistemic function toprovide an organization for this information in terms of various sys-tems of classifications that embody the value of the “cultural lore” ofthat corpus. [...] [In other words] the way we retrieve information isan epistemic activity which allows us to access through the retrievingfilters, how the culture authorities on a piece of information have clas-sified and ranked it within that corpus. With the advent of technolo-gies that automate the functions of accessing and recovering memory,such as search engines and knowledge management systems, meta-memory also becomes part of external memory: a cognitive function,central to the cultural organization of human societies, has become au-tomated—another “piece” of cognition thus leaves our brain in orderto be materialized through external supports

The devices which Origgi refers to are the collaborative content filters, searchengines, microbroadcasting applications, reputation systems, and similar toolsthat go under the label of wisdom of crowds. “Mechanisms that turn individualjudgments in collective decisions/wisdom”, according to the popular definitionby James Surowiecki [54]. Mechanisms that recognize no cultural authorities,but rely on the unattended interaction of large amounts of faceless, dispersed,heterogeneous users in a bottom-up fashion. Emerged in the first place to pro-vide a shortcut to information in the overloaded web, despite only respondingto engineering constraints, these systems have a huge epistemic influence, inthat they shape the way we access the increasing share of the cultural produc-tion of our society that is developed and distributed via the Web. In the nextchapters we shall describe this class of applications in more detail, proposemodels of their functioning and simulatively test them for the possible out-comes in terms of infodiversity. For now, we argue that from a social-cognitive


perspective most of the actual wisdom of crowd mechanisms rely on a commonground that is constituted by:

1. The principle of rating. Implemented, explicitly or not, in the vast major-ity of web applications.

2. A set of algorithms which implement a digital flavor of one ancient socialartifact: reputation.

3. The principle of ranking which is consequent to the above and presidesover the presentation of information.

Basically, the reputation of an item, that is how others value and rate theitem - is the only way we have to extract information about it. This applies- to different extents - to most of the platforms that we will discuss: Google,Youtube, Flickr, Digg, del.icio.us, and others. In all these platforms the infor-mation is presented to the user in the form of a list of items ranked accordingto the presumed relevance or quality, and the algorithms that produce suchan outcome don’t differ much one from the other: all involve the aggregationof an implicit or explicit judgment performed by the community of referenceon a piece of work. An evaluation that is expressed with no explicit referenceto any set of parameters, as Paolucci and ourselves note in [44] and as wediscuss in Chapter 6.

Before moving on let us briefly present an emblematic case of the phe-nomenon described.

2.2.1 Flickr Commons: knowledge goes online, live.

Flickr.com6 is a photo management and sharing platform hosting about 4 bil-lion of photographs7 uploaded by users from all over the world and countingan average of 3 million single accesses daily8. Amateur and professional pho-tographers post their work in the same fashion and network on the basis ofcharacteristics of their photography, interests, or geographical location. Pic-tures are then tagged and often placed into thematic groups on the basis ofcontent commonality. Now one could well argue that the billions of imagesshared on websites such as Flickr serve as a growing record of our culture andenvironment. Here the content selection mechanisms are crucial: Flickr andother online communities based on video/photo/audio are shaped by and or-ganize contents according to the same principles of other Wisdom of Crowdsapplications, that is, based on ranking and (sometimes) the reputation scoreof voters and voted. There is no explicit voting on good pieces of work inFlickr, an (undisclosed) algorithm - instead - ranks the pictures according toa parameter called ’interestingness’ which is a function of the number of viewsand comments a single picture receives in a certain amount of time. The Ex-plore system in Flickr generates a showcase of the best (or "most interesting",6 http://flickr.com7 http://en.wikipedia.org/wiki/Flickr8 http://www.ignitesocialmedia.com/2008-social-network-analysis-report/


as they say) pictures every day/week that is proposed on the homepage. Thequestion here could be: does such feature favour the formation of a common“taste” among the users? Or the propagation of styles/trends via the imi-tation of successful photographers (i.e. those who get frequently featured inExplore)? Other sites, as we’ll point out in the next chapters, reward thosewho post content that encounters the taste of fellow users. This means thatthe “posting activity” of a user can be driven simply by the imitation of themost popular contents in order to raise one’s reputation/karma. What kind ofdynamics does a process like this trigger? How does cultural evolution, opin-ion dynamics, memes circulation take place in such an environment? Do thesesystems move towards conformism, or do they favor innovation? Is there anyroom for those carrying preferences (opinions?) far from those of the majority?What about niches? Our modest prediction is that - far from the enthusiasmthat surrounds the “web2.0” hype - this schema is prone to a number of biases,whose real outcomes on the patterns of information circulation - and hencecultural evolution - are yet to be explored.

Fig. 2.3. Goliath 026 - Most “interesting” image of Flickr for January 9th 2010

Flickr.com recently launched an ambitious project that is very useful toreport here, because it offers a glance into the unfolding process of culturesteadily moving onto the Internet and getting implicitly ranked and orga-nized by collective filtering technologies: the very process described by Origgi.’The Commons9’ is a project launched in January 2008, in partnership with

9 http://www.flickr.com/commons/


The US Library of Congress. It encourages public institutions to upload theirarchives of scanned images to the site in order for users to tag the imagesand make sense of the database. This was a brilliant exploit of the crowd-sourcing techniques triggered by web2.0 for the public benefit. The Library ofCongress holds a database of scanned photographs counting into the millions.This database is little more than useless because the pictures have no metatags, so it is basically unsearchable. Flickr’s users digging into the picturesand adding tags to them are making sense of the collection that otherwisewould be a meaningless bunch of unrelated pictures. The database turns intoa great and handy resource into early photography and also a great documen-tation of the life back then and whatnot. Following the Library of Congressother institutions decided to participate in the project, and now the Commonsdatabase includes photographs coming from public institutions of 18 differentcountries. Now a side effect of this is that, in order to search the database,users will have to pass through the particular ranking system of Flickr de-scribed above. The first pictures to show, for every tag search, will bethose with a high “interestingness” parameter. For example, a searchfor “cathedral” returns 917 pictures, ranging from a still of the restoration ofthe Lund Cathedral in 1860, posted by Swedish National Heritage Board, toan impressive picture of British cavalry passing the ruins of Albert cathedral,Paris, during World War I - from the archive of the National Library of Scot-land. It is extremely unlikely that a user will dig into all of them, one usuallybrowses the first results, just as documented for search engine queries, thuswho is selecting the best ’cathedral’ to look at, is effectively Flickr’s interest-ingness algorithm: a corpus of academic and institutional work is not beingorganized and scrutinized by a committee of scholars following some protocolor some recognized parameters, but “the community”, whatever this means ina context like that of a commercial photography website, is taking care of thedelicate matter.

Flickr Commons organizing and ranking content is the explicit equivalentof one publishing a site on the internet and implicitly getting it ranked andpositioned by Google, and this applies to every single item of knowledge thatgets uploaded on the Internet. An algorithm ranks it and selects what is worthwatching and what is not.

2.2.2 Community judgment, common law and democracy 2.0

We leave the description of Wisdom of Crowd algorithms and their internalsto a later chapter (6). In this paragraph our aim is to advance a hypothesison their origins and antecedents as cultural artifacts. The principles of wis-dom of crowds have basically two forerunners, both rooted in hacker culture.Precisely they draw on the traditional open source principle of “given enougheyeballs, all bugs are shallow ”, a famous saying that resumes the need for asoftware to have its source code available, in order for the community of hack-


ers10 to scrutinize it and find the bugs quicker. Eric Raymond formulates thisprinciple as the ’Linus Law ’ in his famous pamphlet “The Cathedral and theBazaar” [47] in which he compares the open source development model - achaotic and horizontal inclusive community (the bazaar) where theoreticallyanyone has access to the internals of a computer program - to the proprietarymodel of the cathedral, where only a closed circle of developers organized hi-erarchically have access to the code and can therefore correct a software’sbugs. To Eric Raymond the open source model is intrinsically more efficient,exactly because of the ’Linus Law’. This open source approach left the land ofcomputer programming to invade a wider territory the day that Wikipedia11

was launched, in 2001. The free encyclopedia is an experiment of extendingthe ideas of horizontality and anarchism of the opensource movement to abroader scope. The many eyeballs of the millions Wikipedians allow the en-cyclopedia to keep a high level of accuracy and to rival prestigious academicprojects like Britannica [26].

The other branch of the ideology of wisdom of crowd is deeply rooted intoAmerican culture. Reddit, Digg, Google itself claim a direct descendance fromthe American democracy of the origins: from the New England assemblys,where citizens decided of their fate in direct democracy regime, to the anglo-saxon trial, based on common law, with the peculiar role of the popular jury:one is not judged on the basis of an abstract norm conceived, then evolved,then condensed and sculpted in the impersonal rock of Law. The verdict ispronounced by one’s peers who decide whether keep the defendant or expelhim from the social body. The decision takes place on the basis of a sort of“common sense” and the adherence to the norms shared by the social body.And if this holds true for both the Anglo-Saxon and the Roman system, inthe former the social body is physically there and looks at you in the eyes.What are the popularity indexes of reddit, digg or flickr if not a telematicedition of such approach? The consequent passion for ranking and for socialacceptance springs directly from the American model of democracy with itscorollaries of populism and charismatic leadership.

A similar interpretation was proposed in a recent book by the Italianjournalist Carlo Formenti [20], who also refers to the work of Manuel Castellson Internet culture reviewed in the next section.

10 ’Hacker’ equals to ’computer programmer’. Not to be confused with computercrackers or vandals

11 http://wikipedia.org is an online encyclopedia that anybody can edit, withoutthe need of a registration or to document one’s competences. [38]

2.3 “Don’t worry, we’re from the Internets” The persistence of culture in online communities 15

2.3 “Don’t worry, we’re from the Internets” Thepersistence of culture in online communities

After the Court of Stockholm had sentenced The Pirate Bay crew to jail forinfringing copyright laws, The Pirate Bay website12 was still online and active,and a message on the home page calmly and confidently stated:

Don’t worry – we’re from the internets. It’s going to be alright. :-)

What was the stance? In that simple and somewhat cryptic claim the peopleat ThePirateBay were referring to a whole world of sense only shared witha certain community. The “internets” is a bushism13 turned catchphrase usedhumorously to portray the speaker as ignorant about the Internet or abouttechnology in general, or as having a provincial or folksy attitude towardstechnology. To be from “the internets” therefore means to share one partic-ular point of view, ironically watching oneself with the eyes of the outsider.An inside joke, if you like, where the insiders are the inhabitants of the Netthat at the same time reclaim their right to behave according to the customsof their motherland and their ability to come out of the difficult situationgenerated by the ignorance and the arrogance of the outsiders, in this casecopyright legislation and prosecutors. The particular custom at stake in thatcircumstance was the right to freely share information regardless of copyrightlaws and threats from media corporations. An old ’meme’ of the cyberspace,dating back to its first inception (see, for example [7]), that ’infected’ a widerpopulation of netizens over time.

This brings us to the now classic work of Manuel Castells on Internetcommunities and cultures. In his 2001 book “Internet Galaxy” [12] the Cata-lan sociologist enlisted the different communities and cultural tendencies thathave subsequently participated in the growth of the Internet since its first im-plementation. Castells’ claim is that today’s Internet has been shaped by thecommunities that have “inhabited” it along the way and traces of the culturesof those communities persist in every aspect of today’s internet practices. Fol-lowing Castells’ reconstruction, the very community to have had a substantialrole in the present face of the Net is that of the first hackers at Berkeley andthe MIT who put their hands on the first mainframe computers acquired bythose Universities in the late 1960s. To those hackers’ ethic is to be ascribedthe urge for openness and the strong distrust towards top-down regulations ofthe cyberspace that is hardcoded not only into the internet and its protocols,but in the mindset of many of its “inhabitants”. As well as the idealistic idol-atry for the American democracy of the origins that, as proposed in section12 http://piratebay.org. The Pirate Bay is the most popular search engine for

.torrent files, metafiles that contain information for downloading material via thepeer to peer network BitTorrent.

13 George W. Bush spoke of “the internets” twice during the presidential campaignsin 2000 and 2004 - see https://secure.wikimedia.org/wikipedia/en/wiki/Internets


2.2.2, informs the mechanisms of content rating and ranking. The early hackerethic itself, in turn, dues a lot to the libertarian and unconforming culturalatmosphere of the US West Coast in the late 60s and early 70s. The legacy ofthat culture, nowdays, is to be found in the Free Software and Open Sourcemovements that have been driving forces of the development of Internet andits practices14 [36, 29].

Following Castells, right after the hackers came the academic elites of theUS Universities - the first institutions to be connected to each other by theearly ARPANET - for whom the Net was a great opportunity for disseminationof science and long distance collaboration. To one member of this “tribe”, TimBerners Lee, humanity owes the invention of a hypertext system capable ofovercoming the limitations of paper when citing other resources inline: theHyper Text Markup Language which, along with the Hyper Text TransferProtocol, allowed the transmission of hypertextual documents over the Net.And the World Wide Web was born.

In the eighties the user base of the Internet expanded, as almost all thedeveloped world was now covered. Characteristic of this period, accordingto Castells, are the “digital communitarians“. Users of early, text based chatrooms, of Usenet newsgroups and of Multi User Dungeons, online role games.To this period also date the first experiments of political and civic use of thenetworked computer, with the administration of certain cities offering servicesthrough the Net, like Amsterdam’s Digital city [56].

So if TPB crew, all in their twenties, and millions of kids around the worldconsider perfectly legitimate to share works covered by copyright, the reasondates back to the hippies that in the 1970s made internet in a certain way.How does that happen? How is it possible that a cultural milieu dating backto decades still manages to persist in a rapidly changing environment such asthat of “the Internet”?

The work of Indian researchers Ingawale et al. [30] provides a few insightsinto this fascinating issue, from a structural network theoretic point of view.The authors employ agent-based simulation to investigate the issue of culturalpersistence in online communities and it’s well worth an extensive review.The context of the work is that of Wikipedia, the web-based collaborativeencyclopedia built on the basis of voluntary work contributed, for the mostpart, by anonymous users. One reoccurring question about the success ofthe collaborative encyclopedia regards the persistence of the general spirit ofcollegiality and mutual understanding - dubbed WikiLove - that has existedin Wikipedia since the time of its inception in 2001. As the authors notice,the establishment of the norm of WikiLove dates back to the very founders of14 This is true to a great, but most of the times neglected, extent. Think of the

Creative Commons licenses, a cornerstone of the Web2.0 discourse. They areheavily based, and strongly conform to the spirit of the original GNU PublicLicense, the brainchild of the “last true hacker”, as Steven Levy describes RichardStallman in his “Hackers” [36], a comprehensive history of the origins of Internetand cyberculture.


the project, it then spread among the initial population of users which was nowider than a few hundreds of contributors. The near-exponential increase ofthe user base leaves us with the puzzle of how the norm managed to persist,being adopted by the huge number of newcomers.

This work has certain similarities with our own. First of all, to represent a‘cultural norm’ the authors employ the concept of meme, first introduced byDawkins [15], who coined the term to describe a ‘socially transmissible idea’.Ingawale and his co-authors extend the definition arguing that

any idea that can be so transmitted can be a meme, but only thosememes which are universally accepted acquire the status of “culturalnorm”.

They built a network from a dataset of the edits performed on the Wikipediain Cebuano - a language spoken in the Philippines, preferred for the easeof manipulation to a bigger dataset, say, the English Wikipedia. By linkingcontributors that have worked on the same articles and articles that have beenedited by the same contributors, the authors came up with two networks: anauthor network, with the users as nodes and an article network, with eacharticle represented as a node, linked to other articles edited by the sameusers. The networks obtained are used as a battlefield for memes to spread.The study employs concepts from epidemiology, as in the original theory ofmemetics. Memes’ attributes are:

Infectiousness (I) - the probability of a node being infected witha particular meme at each time step. Chance-of-Recovery (R) - theprobability of an infected node from recovering from the infection ateach time step. At every turn, an infected node has a probability I ofpassing on the meme to any of its neighbors and a probability R ofrecovering from the infection.

Therefore the memes have four possible “levels of virulence”, the extreme casesbeing high I + low R (a viral ‘idea’ that catches on quickly but is hard to letgo of ); low I + high R (an idea that is not very infectious, and is also easyto let go of even when once infected). For each of the 4 levels of virulence theauthors perform 25 runs, i.e they infect one article at random at each turn for25 turns with the same setting of I and R. The levels of infection spreadingoutcoming from each virulence level are shown in Fig. 2.4.

The most interesting results regard what happens after the first infection.The authors tried to infect the networks with another meme, and found that

the first meme, once incumbent in the network, is extremely dif-ficult or sometimes impossible to dislodge, even by a highly virulentnewcomer meme. [...] This observed persistence and stability of thenetwork with the old memes, even in the face of the influx of newvirulent competing memes, indicates that a part of the explanationfor persistence of norms does indeed lie in the underlying networkstructure.


Fig. 2.4. Simulation: Representative ‘Successful’ Runs of Responses of theWikipedia Article Network to memes with different values of Infectiousness andChance-of-Recovery. (From [30])

Table 2.2 shows the extreme unlikeliness of a newcomer taking over the in-cumbent meme.

Table 2.2. Descriptive statistics of simulation runs

VirulenceLevel

Infectiousness(I)

Chance ofRecovery (R)

# of runs # of attemptssuccessful atinfecting core

Probab. ofEpidemic that

persists

Success Rate ofNew ‘Challenger’

Meme withVirulence Level 4

4 80% 20% 25 25 100% 0%

3 20% 20% 25 9 36% 4%

2 80% 80% 25 4 16% 8%

1 20% 80% 25 0 0% 100%

The reason for this resilience, argue the authors, is to be found in thestructural properties of the networks emerging from the interaction patternsobserved. Such networks show two key properties: (a) a scale-free distribu-tion of links and (b) the assortative mixing of nodes.


• (a) The Scale Free distribution observed in many real networks is charac-terized by the presence of a small number of nodes with a large numberof links and a large number of nodes with a small number of links [4].This distribution of links is observed both in the user and in the articlenetworks.

• (b) Networks are said to exhibit assortative mixing (or positive correla-tion) if nodes of a given degree tend to be attached with higher likelihoodto nodes with similar degree. Assortative networks have been found topercolate15 more easily and are more robust to removal of their highestdegree vertexes [41]. This property is found in the article network.

The network-theoretic explanation of the persistence of WikiLove leans par-ticularly on the fact that

the high-degree vertices will tend to stick together in a sub-network orcore group of higher mean degree than the network as a whole. Percola-tion occurs earlier within such a sub-network. The core group of an as-sortatively mixed network thus fork a “reservoir” for the virus/meme,sustaining an epidemic/culture even in cases in which the network isnot sufficiently dense on average for the disease to persist. In an ex-ponentially growing online community, a large number of new users,with very few interaction ties are added every time period. In spiteof this, if the “reservoir” has been formed, the original cultural normspersist and eventually are adopted by the newcomers, as a result ofrepeated interactions with the core group of high degree nodes.

The authors conclude their analysis with a very interesting remark on the na-ture of the network structure of Wikipedia and its sustainability. Comparedto studies of other co-authorship networks, like academic publishing or Broad-way musicals, the Wikipedia co-authorship network is, in a way, more elitist,in that

the majority of the nodes lie outside the [...] dense but small ‘giantcomponent’ [made of] high degree nodes [that] preferentially connectalmost exclusively to other high degree nodes.

Figure 2.5 clearly shows that the fraction of the nodes in the giant component -the ‘closed club’ responsible for the positive externality of providing stabilityand the much discussed persistence of cultural norms, and consequently, areliable governance mechanism - is slowly, but steadily decreasing.

The interaction ties by which the inner core of nodes controls andcoordinates Wikipedia are gradually getting strained

meaning that a breaking point of social production may well be lurking aroundthe corner. What will happen then? Will Wikipedia explode in an apocalypse

15 Percolation is the formation of a giant component – the largest connected chainof nodes.


Fig. 2.5. Average component size and percentage of nodes in the giant component.(From [30])

of uncontrollable flames? Will a new norm establish at the expenses of thosewhich have regulated the community so far? Will a fork in the project takeplace?

Whatever the fate of Wikipedia and Wikilove will be, the re-search and the findings of Ingawale and colleagues elicit a puzzlingquestion: could the dynamics within Wikipedia be representing inscale those of the entire Web? In other words, could the fascinatingnetwork-theoretic structural explanation of the persistence of cul-tural norms in Wikipedia proposed above be extended to explainthe persistence of other practices and customs, inherited from theearly days of the Net and still in full effect decades later, as ar-gued at the beginning of this chapter? And if so, what will theirfate be, after several subsequent endless septembers16 populatedthe cyberspace with newbies who rarely relate to the old timers,who in turn are often self-segregating to certain sectors of the Net?Could the commercial platforms - like Facebook, which introducedthe norm of using one’s real name and surname, while “the Internet”had always been proudly anonymic, or pseudonymic - be aggregat-ing a new giant component that will “outlaw” those old timers who

16 Eternal September (also September that never ended) is a Usenet slang expressioncoined for the period beginning September 1993. The expression encapsulates thebelief that an endless influx of new users (newbies) since that date has contin-uously degraded standards of discourse and behavior on Usenet and the widerInternet. Usenet originated among universities, so, every year in September, alarge number of new university students from the Northern hemisphere acquiredaccess to Usenet, and took some time to acclimate themselves to the network’sstandards of conduct and "netiquette". After a month or so, these new userswould theoretically learn to comport themselves according to its conventions.September thus heralded the peak influx of disruptive newcomers to the network.(https://secure.wikimedia.org/wikipedia/en/wiki/Eternal_September)


still stick to funny nicknames, and look at them as weird and freakypeople?

3

The Meme-To-Web model

Let us consider a society as that described in Chapter 2, with the featuresresumed in table 2.1: a multitude of small, independent content producerspublishing cultural artifacts to a publicly accessible virtual space. The arti-facts (in the form of music, art, photography, writing, poetry, ...) carry theproducer’s beliefs and knowledge, and are connected to each other on the basisof some content commonality. Let us also imagine that no direct interactionamong agents exists in this society and all the communication takes placethrough the artifacts1: some produce them, all the population can access and“enjoy” them, thus internalizing part of the content they carry. The “accessto knowledge” (i.e. to the network of artifacts) takes place via the media-tion of one or more technologies resembling web-based epistemic algorithms -search engines like Google’s PageRank, popularity indices like in Flickr.com,collaborative content filters.

It is the society we reproduce in the Meme-To-Web2 model: an idealtypicpeer-production society, with no influx from outside the system. The questionwe ask is: if such a society were to be - as one day it might be, if the dreams ofBenkler and others come true - would it be an infodiverse society? that means,will a lot of heterogeneous information be available to users, with chances todiffuse? Will a meme held by a minority of the population be able to spreadand conquer the rest?

1 An extreme assumption, we agree. It reminds of certain dystopical criticism thatresurfaces from time to time, warning against an atomized society in which hu-man interaction is completely replaced by computer based communication. Theextreme assumption however is useful for our investigation in that modeling aphenomenon in its most extreme possible form helps surfacing the nude dynam-ics in their crudeness. In this sense the Meme-To-Web model falls in the categoryof ideal-type models, as formulated by Nigel Gilbert in [24], i.e. one in which somecharacteristics of the target are exaggerated in order to simplify the model.

2 The complete annotated code is downloadable from http://dl.dropbox.com/u/501682/internetz_019b.nlogo

24 3 The Meme-To-Web model

To try to give an answer to this question we have to look at the tools thatmediate the access to content. They are crucial, because they select whichbeliefs, chunks of knowledge, ideas will more likely spread in the population.Will something get lost? In other words, will the attitude of the internet to-wards fringe markets popularized by Chris Anderson [2] apply to fringe ideasas well? Which, of the content filtering algorithms, will let each of these cir-cumstances take place? Content filtering applications appeared subsequentlyin the history of the Internet, one at a time. The primeval Internet had nomeans to filter content other than the immanent categorization produced byhyperlinking. Then, there were simple keyword-based search engines, such asAltaVista. Later Google conquered the search engine market with its superiorPageRank algorithm, then reputation systems and collaborative content fil-tering sites were born. Finally, we now have social networks, which let contentflow according to established personal links. Is there a pattern - in this shortby extremely dense history - that can be recognized, with relation to infodi-versity? Are we, for example, moving from more ’plural’ platforms, allowingless popular ideas be accessible, to platforms that filter out dangerous stuff?Or maybe the other way round is the case?

To test some of these claims we now introduce the Meme-to-Web model.Before describing it in detail we briefly introduce the methodology employedand review a few attempts made at modeling some of the activities related tocontent production and fruition in web based contexts.

3.1 Agent based simulation

Agent-based modeling is a new analytical method for the social sciences thatis quickly becoming popular. It enables one to build models where individualentities and their interactions are directly represented. In comparison withvariable-based approaches using structural equations, or system based ap-proaches using differential equations, agent based simulation offers the possi-bility of modeling individual heterogeneity, representing explicitly agents’ de-cision rules, and situating agents in geographical or other kinds of space [24].With Agent-based Social simulation (ABSS) we generate artificial societies ofautonomous and heterogeneous agents able to interact, communicate and co-operate. Models of these societies are usually, but not necessarily inspired byreal world societies that can not be easily analyzed, especially when the objectof investigation is a complex and emergent phenomenon. Local interactionsamong agents produce the macro-phenomenon that is investigated throughthe implementation of local rules followed by the agents [25]. Agent-basedsimulation can be paired with computational modeling of cognitive systems,thus leading to agents endowed with internal representations (beliefs, goals,intentions) that drive agents’ actions in the world [13]. One could ask whetheragent-based simulation is the most effective tool to inquire into a domain likethat of internet based technologies and practices. It is a domain where, after

3.2 Related works: models of blogging behavior 25

all, a wealth of empirical data exists, useful for both quantitative and quali-tative research. We believe that the simulative approach makes sense and isperfectly suited for the problems at stake: understand how the technologiesof content filtering that have appeared subsequently in the history of Internetaffect the shape and richness of the informational environment. In our viewthe studied setting is a complex system particularly characterized by constantfeedback loops between production and consumption of goods and the bestmean for observing such dynamic is to reproduce it in vitro. Moreover, oneaim is that of discovering whether one system is inherently more pluralistthan the other and no other method could document the functioning of eachsystem in isolation as a computer simulation. We plan to exploit, however, thevast amount of data available in order to validate the model, to gain insightson whether the dynamics springing from the model are plausible or not.

3.2 Related works: models of blogging behavior

While the model we propose is abstract and doesn’t refer to a single specificphenomenon, the concepts and the approach that we employ are sometimesfound in internet research and blog models. A couple of recent works try tomodel separate aspects of the browsing and publishing behavior of web users.These are the only attempts that we are aware of at simulating the humanbehavior in blogging contexts and are employing an approach similar to ourown.

In a recent paper, the authors of [39] draw on a work that we describe be-low, in section 5.1 and propose an agent based model of the browsing behaviorof users which happens to include some insights also present in our model.With their “ABC” model3 the authors aim at building a realistic model ca-pable of explaining traffic patterns observed in real world. They captured thebrowsing behavior of college students using faculty computers and built theirmodel starting from three assumptions:

• agents maintain individual lists of bookmarks (a non-Markovian memorymechanism) that are used as teleportation targets.

• agents can retreat along visited links, a branching mechanism that alsoallows us to reproduce behaviors such as the use of a back button andtabbed browsing.

• agents are sustained by visiting novel pages of topical interest, with adja-cent pages being more topically related to each other than distant ones.

The model assumes a scale free network of sites and lets users surf back andforth in such network, with the ability to teleport to sites not directly linkedusing a bookmarking feature. Agents are endowed with an amount of energythat gets consumed when browsing non relevant sites and it’s increased in

3 ABC stands for “Agents, Bookmarks and Clicks”


the opposite case. Their conclusion is that the model provides more realis-tic assumptions on users behavior than, for example, those “hardcoded” inthe PageRank algorithm, which simply assumes that users random walk thenetwork following links.

In [19] the ZeroCrossing model is presented. It is a model aimed at repro-ducing both the topology and the temporal dynamics of the blogosphere, withassumptions on reading and posting behavior of bloggers. Of particular inter-est is the “inter-posting time”, defined as the time between two consecutiveposts of the same blogger. The authors base it on zero-crossing of a randomwalk on a discrete line, as shown in Figure 3.1, claiming that it reproducestemporal patterns observed in real world blogs. At each time step an agentcan be in a state represented by an integer with two possible transitions: withequal probability the publisher either adds or subtracts 1 from his currentstate and publishes an artifact when his state is 0 (i.e., “in the mood for pub-lishing”), thus performing a random walk on a line. We chose to adopt this

Fig. 3.1. The possible agent’s states in the Zero-Crossing model. A post is publishedonly when the agent is in state 0.

same approach in our model (see section 3.3), being it, to our knowledge, theonly model of this particular aspect of blogging available. The decision to linka post to another blog, then, is a mix of preferences and random choice. Themodel is set to produce a scale free network of blogs, the observed shape ofthe blogosphere.

The authors of [31], to conclude this brief survey, propose a generativemodel of the blogosphere based on a mix of preferential attachment towardshighly connected nodes and random surfing, again in order to produce a per-fect power-law indegree distribution.

3.3 Meme-to-Web. An abstract, generative, agent-based model of peer-based content production and access 27

With respect to these works, our model tries to capture - at the same time- the dynamics of content generation, those of access to content, the feedbackloops between each other, as well as the influence on both exercised by thepeculiar infrastructures of the web, such as search engine algorithms.

3.3 Meme-to-Web. An abstract, generative, agent-basedmodel of peer-based content production and access

The Meme-To-Web model implements an artificial society characterized bypeer-production and consumption of cultural artifacts. In the MtW model asubstantial share of the population produces cultural goods (artifacts4) whichare made available to the whole population at no charge. In our abstractimplementation each artifact is no more than a set of memes, that are - in aneglected, but still useful metaphor [15] - basic items of knowledge existing inthe form of atomic signs contained in agent’s minds and embedded in culturalartifacts. The model, therefore, contains an hypothesis on content creation,namely that information artifacts are:

• created by associating basic idea components (memes) that populate elec-tronic artifacts and agents minds, both acting as containers. While mindsare dynamic containers (the memes contained change in time as a resultof agents actions), artifacts - once instantiated - cannot be modified;

• connected to each other on the base of memetic content

The circulation of ideas and thought (i.e. of memes) takes place through thedistributed production and circulation of the artifacts, which - once createdindividually - are stored in a separate network of directed links between ar-tifacts. The access to the artifacts in the network is mediated by a set ofalgorithms modeled after filtering technologies at work on the web.

In the version of the model presented here, artifacts are accessed fromusers in different ways according to experimental conditions:

• RandomWalk : simply random-walking the graph by following links froman artifact to another (discussed in chapter 4)

• PageRank : searching for known memes via the mediation of PageRank(Chapter 5)

• Reddit: (Chapter 6) Reading artifacts indexed on the basis of the commu-nity rating they received.

In the model, agents search and consume information and the result of hy-pothesized user activity gives shape to the actual network of artifacts, whichreflects again what was inside the minds of publishing agents.

4 An artifact can be thought of as a song, a blog post, a piece of video, a photo-graph: any reproducible aggregated item of knowledge, such as those producedor transmitted via a computer on the Internet


3.3.1 Model internals

Fig. 3.2. The M-t-w model

The agent’s mind

This part of the model corresponds to what, in behavioral sciences, is oftencalled a black box system - the set of concepts in the content creators’ mindsand the processes that allow them to select, assemble, publish and dissemi-nate them. Finding the appropriate level of description for the agents’ mentalprocesses is not an easy task. To keep the model manageable, we assume forthe present that these concepts can be represented by atomic, meaninglesssigns. Under this assumption, it makes no sense to build processes aroundthese memes that would not be processes of random selection. This approachis not novel; it can be compared to the tradition of the bit string culture mod-els, starting from Axelrod’s model on evolution of culture [3]; we have morerecent and advanced examples also in [55], where symbols are combined toform models of Wikipedia pages. With respect to these approaches, we takean even simpler position, with “empty” signs that travel in the communica-tion space. In further versions of the model, we could add relations between


memes; for example, to explicitly structure competing memes, in the simplestway as ideas that contradict each other: eating fish is good, eating fish is bad.

Agents are built with a simple cognitive infrastructure, consisting in threecontainers:

1. ’beliefs’ contains memes accepted as true2. meme-memory contains memes encountered during surfing3. artifact-memory contains artifacts encountered during surfing

We model the belief base as a "meme store" (see Fig.3.2). When a new artifactis encountered the memes contained are, at first, stored in a limited memory,and the artifact itself is remembered for a limited time. The set of memescontained in an artifact can be either coherent (in our view: containing atleast a meme already believed by the agent), or containing completely newinformation5. In the first case, according to [11] there is a large probabilityof acceptance of one or more of the other memes contained in the artifact.In the latter case the new memes will have a smaller chance of being remem-bered and will have a harder time in being incorporated in the belief base.We designed the architecture keeping these simple principles in mind, so ifthe artifact contains at least one known meme it will be retained in mem-ory according to a probability sticks-in-mem, while for completely unknowninformation the chance of retaining is set to sticks-in-mem/2. In this firstimplementation we have a simpler mechanism for belief manipulation, though:we use reiterations (the times a certain meme has been encountered) asa measure of the probability that the meme will be embedded in the agent’sbelief base, as in classical cognitive theory (see [23]).

Memes

A brief digression has to be made into the sad history of memes: a powerfuland fascinating metaphor that has failed its promises. Richard Dawkins [15]proposed that the meme is to culture what the gene is to biology, a replicator –“a unit of information with the ability to reproduce itself using resources fromsome material substrate”– that serves as a basis for the transmission of culture.Any idea that can be so transmitted can be a combination of memes. It is afascinating hypothesis, the idea of mirroring what we know of the interplaybetween the genetic code and evolutionary pressure. Computationally it wouldhave opened the field of epidemiology to that of cultural transmission

We will just work out the model as if these unit of knowledge existed,hoping to draw an indirect confirmation of our assumption by the plausibilityof the network that we are going to grow. In this work we take this as aworking hypothesis, that is, we are not pretending that these knowledge unitsreally exist - it is the object of a heated debate that we don’t plan to enterhere. Memes are still to be found in the wild, and they don’t have any obvious

5 In this version we don’t have conflicting memes


physical counterpart, nothing even vaguely comparable to the DNA for thegenes [17]. Another critique comes from semiotics, (e.g., [16, 33]) stating thatthe concept of meme is a primitivized concept of Sign. Meme is thus describedin memetics as a sign without its triadic nature. In other words, meme is adegenerate sign, which includes only its ability of being copied. Accordingly,in the broadest sense, the objects of copying are memes, whereas the objectsof translation and interpretation are signs.

Simulation cycle

The main simulation cycle consists of the six subsequent functions describedin detail below: Initialization -> Publishing -> Linking -> Exploration ->Internalization -> Corruption. Table 4.1 shows a list of parameters and theirdefault values.

Initialization

Agents are created with an initial random number of memes, a Poisson distri-bution centered on NMemes + 1, plus an average readingCap, again Poissondistributed in the population, representing the average number of cultural ar-tifacts “consumed” by each user at each time step. A pct-publishers fractionof the agents is then attributed with publishing abilities. The Poisson distri-bution produces a population of individuals characterized by mostly similarabilities with small fractions deviating in a sense or in the other.

Publishing

The users with publishing abilities have the chance of producing one artifacteach and endow it with a limited number of memes (a maximum of 0.20*total) picked randomly from {beliefs}. For determining the publishing fre-quency we adopt the zero crossing blogging model [19] as described in section3.2: at each tick, a publisher is in a state represented by an integer with twopossible transitions: with equal probability the publisher either adds or sub-tracts 1 from his current state and publishes an artifact when his state is 0(i.e., “in the mood for publishing”), thus performing a random walk on a line.According to the original authors “this mechanism generates bursty publish-ing activity: the publishing time-stamps are exactly the zero-crossings of arandom walk (Brownian motion)” and produces average inter-posting timesempirically observed in blogs.

Linking

The newly created artifact is then linked to some of the artifacts stored in{art-memory} (i.e. artifacts that the creator had encountered in previous


surfing). The assumption we make is that the linking happens on the basis ofcontent commonality: once an artifact is created, the agent creates a directedlink towards known artifacts with the most similar meme set. A small chanceof reciprocation exists: with probability avg-recpr a link from the targetartifact is drawn to the new artifact. At the same time the author links thenew artifact with one of her own other artifacts picked from {creatures}.

Exploration

In this phase the user employs the tools at his disposal to gather informationand to explore the network of artifacts

• At the beginning, he will perform a RandomWalk on the network of arti-facts: the agent picks a random artifact from his memory and follows theoutgoing links for readingCap steps, reading artifacts. We will discuss theoutcomes in Chap.

• In the PageRank condition the agent selects a meme from {beliefs}and performs a search for artifacts containing it. The artifacts received inresponse are the top 10 artifacts that embed that meme, ranked accordingto their pagerank value 0 ≤ p ≤ 1, computed for every artifact at eachtick. A number of such artifacts, no larger than the user’s readingCap,will be read by the user.

• In the Reddit case users rate the artifact read on the basis of memeticsimilarity with their belief base6, when reading the users are presentedwith a list of the best ranked artifacts and will read an amont of artifactsequal to readingCap

• The Hybrid algorithm contains portions of all the above strategies. Theamount of artifacts to be read, readingCap, is divided among the threeabove access. The user selects a meme from {beliefs} and performs asearch for artifacts containing it - just like in the PageRank case. Thenumber of artifacts consumed in this case is only readingCap / 3; the restof the reading is done following links from one of the artifacts consumed,like in the RandomWalking case, and reading some artifacts selected byother users, like in Reddit. We consider this strategy as the most realis-tic one, assuming that the browsing habit of real web users consists ofthe combined exploitation of the searching facilities and of the hyperlinkstructure of the web.

Memes contained in that artifacts read will be stored in the user’s memoryaccording to the described constraints, should the meme be already there itwould gain one reiteration.

6 This case contains a strong assumption on the activity of rating in content filters.One of the puzzling aspects of content filters is that the reference of the ratingis unclear, or liquid. We chose to base the behavior of our agents on memeticcommonality and discuss the reasons for this decision in Chapter 6.


Retaining and corruption

The artifacts visited are stored in the limited {art-memory}. Similarly ev-ery time a user encounters a certain meme it gains a reiteration point in{memory}. As stated above we use reiterations as a measure of the proba-bility that a certain meme will be embedded in the agent’s belief base, soafter a given number of timesteps users update their belief base integratingmemes with most reiterations. Symmetrically, memories with fewest reitera-tions are removed. The same applies to beliefs stored in {beliefs}: they gainconfidence when encountered often and periodically those with the lowestconfidence score are dropped from the belief base.

3.3.2 Implementation

The model was implemented in NetLogo [57]. We represented memes, artifactsand users as agents (turtles) and the relations among them as typed links.Agents create artifacts, link them to a fraction of the memes in their belief-base, and connect them with other artifacts, generating the directed networkof artifacts (the “noosphere”). Agents then explore the network of artifacts,reading some of them, and store in memory a subset of the memes contained.The outcomes of agents’ actions are represented by three networks:

1. the directed network between artifacts, whose links are created by pub-lishers, representing a concrete structure to be explored;

2. the weighted network connecting agents to memes, representing the dis-tribution of memes in the agents’ minds and the confidence towards eachmeme;

3. the network connecting artifacts to memes, representing the ideas behindcontent production;

At each tick the pagerank value ranging 0-1 is calculated for every artifact.The PageRank code employed in the model is an adaptation of [53] in its“diffusion” version.

Table 3.1 and table 3.2 summarize the main attributes for agents and linksused in the model.

Why Netlogo?

The choice of NetLogo for the implementation of the MtW model has theadvantage of allowing easy code publication and reusing. We hold this asfundamental in the effort to push social simulation towards maturity. Socialsimulation, celebrated by Axelrod in a renowned paper [3] as a new way ofdoing science, may hold the keys to a new understanding of social science.But the language of the new science is still to be found; simulation cannot yetdescribe its algorithms in the detail needed for replication within the spaceof the scientific paper. A simple but rich language as NetLogo allows us to


Table 3.1. Agents and variables

Agent Attribute Description

user ispublisher whether a user is also a publisher

readingCap artifacts a user can read at each step

{creatures} list of artifacts produced by a publisher

{beliefs} memes in belief base

{meme-memory} list of memes encountered in surfing

{art-memory} list of artifacts encountered in surfing

status implements randomwalk for publishing

artifact pagerank pagerank value

pageViews hits received

creator the user that produced the artifact

{memes} memes embedded

meme

Table 3.2. Links

Links connecting description attributes

memories minds – memes Represent the memory of users reiterations

beliefs minds – memes Represent the belief base of users confidence

weblinks artifacts – artifacts The networked noosphere

memetic artifacts – memes

share the code that implements the model, with the actual hope that this codecould be read by interested parties, could be used for replication, modification,and education; to the contrary, the an ordinary programming language tendsto contain the potential audience to the intersection of social scientists andtechnologists - a rather scarce set.

Of course there is a price to pay; with the code we have produced, we areunable to obtain networks with a size comparable to the ones observed in theweb, as documented in Chapter 4. Still, we think that our networks are ableto reproduce the essential features of their larger sibling from observations;furthermore, conversion to an ordinary object oriented language for perfo-mance increase is not a difficult task once a working implementation exists.In short, we have chosen readibility, manageability and fast prototyping overoptimization and speed; ideas over technique.


3.3.3 Model falsification

The While we cannot expect quantitative agreement with real network, thereare several basic questions that we can use to falsify the model. In this work,we focus on the most These are

• memes make a difference. The random strategy produces a different effect.• memes produce a more plausible network, that is, a network that shows

structural characteristics of an actual network.• memes drive the dynamics of exploration, when significant

To sum up, we could soberly define the Meme to Web (MtW ) model as an ab-stract, generative, (slightly) cognitive, agent-based, content-drivenmodel of the peer-production web.

3.3.4 Strengths and limits of the model

Apart from the aforementioned reliance on a spurious implementation of thenotion of “meme”, the model has several possible improving points.

First of all, as of the present version of the model, the memes exist in-dependently one from the other. We are missing the memeplexes, i.e. stableconfigurations of memes that appear in chain, or together. As of ideas thatmost of the times co-occur in a population. Memeplexes will be implementedin a future version of the model, one interesting thing to investigate will bethe possible emergence of memeplexes within the system as is, meaning thatmemeplexes won’t necessarly be designed in, but instead might be “seen hap-pen”.

Simmetrically, we are missing the possibility for incompatible memes,which is memes that can’t be accepted as true at the same time. This could bea huge theoretical pain, as - sticking to the original formulation by Dawkins- there are no such things as incompatible memes. However for the definitionthat we have given in the model, that contemplates the existence of memes“accepted as true” by an agent, we have to hypothesize the possibility of in-compatible memes (“god” / “no god” case). This should be of particular usefor exploring the effects of PageRank, when searching for a particular memeone can be presented with artifacts carrying the exact opposite. (the Barbiecase, see chapter 5).

Another general critique to the model is that, at present, it misses cognitiveelegance and richness, in particular some assumptions seem way too arbitraryand need polishing.

However the model is very easy to manipulate, the internalization/corruptionrules can be adapted to a better theoretical grounding and, in addiction, car-ries one important feature in that it reproduces the feedback loops be-tween content consumption and production, something which, to thebest of our knowledge, has not yet been tried and could be of some importancefor reproducing and enlightening.


3.3.5 What to look for and why

The ultimate aim of Meme-To-Web is to help understand the ways contentfiltering algorithms affect the circulation of ideas and thought in a peer pro-duction society. For achieving this, we need a measure of the richness andvariety of the information environment. And we found our measure out there,laying in the land of analogies...

“Biodiversity is the degree of variation of life forms within a givenecosystem. Biologists most often define biodiversity as the "totality ofgenes, species, and ecosystems of a region”7

Having abundantly exploited the biological analogy with the memes, wethought of not stopping there and introducing the notion of infodiversity,defining it as the quantity of memes circulating in an infosystem. Avariety of measures for information diversity could be employed, indeed [32]:for example one still notably used, 60 years after its initial formulation is theShannon index [51]. However, for our preliminary results, in order to gaina quick impression on the dynamics at work we will stick with the simplestpossible measure, that is: analyze how many memes keep circulating in thesystem after a certain period and their concentration rate.

It is useful to notice that we are implicitly assuming that more diversity isintrinsecally preferable to less diversity in an infosociety. Such a claim could bedeemed as obvious - and it is so in many contexts, expecially those critical withthe alleged homogeneization of information8 - but in fact it is not. Cognitiveliterature, for instance, has plenty of evidence that a system with a greatdeal of incompatible and inconsistent information is prone to a number ofdysfunctions (see for example [52]). We won’t dig into the debate, but considerthis matter yet one more important theoretical caveat posed on the model.

7 http://en.wikipedia.org/wiki/Biodiversity8 For an enlightening example of this approach here is a quote from AdbustersMagazine

Cultural homogenization has graver consequences than the same hairstyles,catchphrases, action-hero antics and video clips propagated ad nauseamaround the world. In all systems, homogenization is poison. Lack of diversityleads to inefficiency and failure. Infodiversity is as critical to our long-term survival as biodiversity. Both are bedrocks of human existence.(https://www.adbusters.org/magazine/90/ecology-mind.html)

4

Build a network of artifacts, (random) walk onit

We let Meme-To-Web run for 2000 steps with the parameters resumed in Table4.1. Content producers are set to a quarter of the population, thus creating asituation of peer production. First thing we try is to allow our publishers tocreate their artifacts, link them together and let all the agents surf the networkemerged reading content. We will examine the shape and properties of thenetworks that emerge from this kind of access, one that exploits the hyperlinkstructure of the network, without any further mechanism of content selection:plain surfing. We keep this surfing model as a baseline condition, in the nextchapters we will add two more content filtering applications, PageRank andcollaborative filtering, and compare the outcomes in terms of network shape,access patterns and infodiversity.

Table 4.1. Global parameters

Parameter value

initialUsers Initial # of users 200

nMemes number of memes circulating in the society 1000

pct-publisher share of users who are also publishers 0.25

NArtefacts number of artefacts produced at each turn 1

AvgReadingCap average number of artifacts read each turn 5

sticks-in-mem probability that a meme is remembered 0.85

memes-in-art share of a publishers memes in each artifact 0.20

conf-loss rate of confidence degradation if memes not reiterated 0.2

avg-recpr chances that a link to an artifact is reciprocated 0.05

max-memes-bel limits of the belief base 50

38 4 Build a network of artifacts, (random) walk on it

4.1 Publishing phase: network growth and similarities

We examined the network of artifacts emerging from the publishing activity,focusing on the shape of the network and the degree distribution, and com-pared our artificially-grown network with a reference, the Stanford web graph1

[34]. This is a fairly extensive collection consisting of a web crawl performedstarting from the University of Stanford domain, retreived in 2002.

The network produced by publishers has 26969 artifacts (nodes) and144576 edges (links). The distribution of inlinks has an ample zone follow-ing a power law with exponent -1.80. This is consistent with the well knownfindings of Barabasi at al. popularized in [5] with regards of the web: a smallnumber of artifacts holds the great majority of incoming links, while most ofthe artifacts are hardly reachable, because connected only to a few other ar-tifacts. The comparison with the Stanford dataset shows a striking similaritywith regards to this feature: Table 4.2 and Figure 4.1 compare the two net-works. In particular Fig. 4.1 presens the degree distributions on a log-log scaleshowing the extremely similar shape of the real network and the simulatedone.

Table 4.2. Network properties

Nodes Edges Power law exp.

LinkFollowing - simulated 26969 144576 -1.80

Stanford - dataset 281904 2312497 -1.71

Our purpose is not of accurately reproducing a specific network; rather,we focus on the different mechanisms of growth, and we try to discern whatkind of general network shape the theory produces. The growth mechanismproposed in this work, even if highly simplified with respect to the actualcognitive processes that lead to peer content production, is intricate enoughto escape direct mathematical analysis, but it can produce behavior that wecan analyze - reaching understanding by growing, à la Epstein[18].

1 Data available on http://snap.stanford.edu/data/web-Stanford.html

4.2 Memes and minds: looking for infodiversity 39

● ●●

●●

●●●●●●●●●●●

●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●

●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●

●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●

1 10 100 1000 10000

1e−

051e

−03

1e−

01

Index

stan

f.dg

● Stanford datasetLinkFollowingpower law fit

Fig. 4.1. Cumulate node density vs. number of links on a log-log scale for the sim-ulated network of artifacts and an observed distribution. The Stanford network, aweb 1.0 collection, resembles the simulated network generated by the RandomWalkalgorithm: they both have a power law zone (with very similar slope, -1.8 for Link-Following, -1.71 for Stanford) with a sharp cutoff (B).

We present The network built on the basis of random surfing is in the initialtrait a scale-free network with the distribution of inlinks following a power-lawwith exponent -1.8. We compare this distribution with a well-known dataset

4.2 Memes and minds: looking for infodiversity

Let us now take a look at the distribution of memes in the minds of agents.We ran an ad hoc simulation with 5000 memes circulating in the society2 and500 users. After 1500 ticks the amount of memes still present in the systemwas of 4418, with a rough 8% of memes (382) expelled from the system. At thesame time one meme had managed to diffuse in the 75% of the population, 9in the 25%.

2 Due to the randomic initial distribution only 4800 of the memes were effectivelyrepresented in the minds of agents

5

PageRank and its discontents1

Search engines are the oldest technology for information filtering at work onthe web. They are so ubiquitous and at this point essential to be often per-ceived as embedded in the WWW fabric. They have been scrutinized andinvestigated extensively (ref.) from various points of view, including the pri-vacy issues that can derive from having one’s search log stored and used forpersonalized advertisments (Ippolita).

The PageRank is the best known and most effective search engine algo-rithm to date. It constitutes the basis of the most successful search engineso far, Google, developed by then Stanford students Sergey Brin and LarryPage [10]. From its introduction Google has substantially changed the way weapproach the web, turning a bunch of unrelated data into the closest thingto a oracle that humanity knows today. Thanks to the enormous success ofits search engine Google’s importance (and capitalization) has raised to an alltime maximum level for a “new economy” company. Nowdays its ubiquity isboth fascinating and somewhat unnerving, with the almost total control onan extravagant amount of user’s personal data - including browsing habits,personal mail, social network data - that it holds thanks to the successful webbased services offered.

While the algorithm is not public its basic functioning is documented wellenough to build a simple model of it and test it in MtW. PageRank interpretsa link from page A to page B as a vote, by page A, for page B. But PageRanklooks at more than the number of votes or links received: it also analysesthe page that casts the vote, so that votes cast by pages that are themselves’important’ weigh more heavily and help to make other pages ’important’2.’Important’ here stands for : successful at attracting hyperlinks. Every singlewebpage is assigned a pagerank value, which is a function of its indegree, i.e.the number of incoming links it holds. In computing pagerank every incoming

1 Excerpts from this chapter were presented at the Third World Congress of SocialSimulation [46]

2 http://www.google.com/corporate/tech.html

42 5 PageRank and its discontents

link itself counts according to the pagerank value of the originating page.When returning results from a user’s query, Google will show the pages withgreater pagerank on top. The system is similar to an election where the votescount differently according to who the same: those originating from prestigiouselectors, that is those with many incoming links, count more. At the end,for each query, the page with more ’votes’ will be featured on top of thesearch result page. What is strikingly interesting of this approach is thatPageRank implicitly does what Reddit, Digg, Slashdot and other collaborativefiltering sites that we will examine in the next chapter do explicitly: it ranksinformation on the basis of the ’approval rate’ gained by the community ofreference, in this case expressed in the form of outgoing links from other sites,and makes those votes count in the evaluation of others’ work, i.e., whenlinking another website.

The mechanisms of emergence of web2.0 - those which, according to com-mon sense, put collective intelligence at work selecting the best contributions -are based on rating and ranking procedures. Ranking algorithms, in turn, arebased on a very broadly intended principle of democracy. He who has morevotes (actual votes, pageviews, friends in the profile page, number of visits) isrecognized as "successful" and ends up on top of the list. This system is proneto a number of bizarre errors. For example, when linking to a fraudulent webpage as a warning to other users, PageRank will count the link as a positivevote3.

Where does this approach take? Benkler has his say on this, although im-plicitly. Praising the “culture both plastic and critical” encouraged by Internetuse, as a glamorous example he reports the case of a Google search for Barbiedolls. Such search, among the top results, features sites both appraising (orselling) and criticising the doll. According to Benkler thanks to the Internet,and in this case also to Google,

[...] the contested nature of the doll becomes publicly and everywhereapparent, liberated from the confines of feminist-criticism symposiaand undergraduate courses [. . . ]

What Benkler is praising in that segment is the peer-production approach inits whole. Dispersed individuals produce content of very different nature overa certain issue, publish it thanks to accessible technologies and such contentbecomes instantly available to large audiences that can discover “the contestednature” of the issue. Is it true?

3 The practice known as ’Google bombing’ relies on this particular characteristic ofthe search engine: linking a page using a particular keyword associates the targetpage to that keyword, and the more pages hold links pointing to that particularpage with that particular keyword the more prominent the association will be inthe search result list.

5.1 Googlecracy, or the “search engine bias” 43

5.1 Googlecracy, or the “search engine bias”

It has been advanced that search engines, far from the Benkler’s optimism,bias the traffic of users according to their page ranking strategies, and somehave argued that they create a vicious cycle that amplifies the dominance ofestablished and already popular sites, with the loop shown in Fig.5.1. A phe-nomenon usually referred to as “search engine bias”, or with a more imagefuldefinition “Googlearchy”[40].

Fig. 5.1. The search engine bias as illustrated in [21]. A. Page i is “popular” in thatit has many incoming links and high PageRank. A user creates a new page j. B. Theuser consults a search engine to find pages related to j. Since i is ranked highly bythe search engine, it has a high probability of being returned to the user. C. Theuser, having discovered i, links to it from j. Thus i becomes even more popular fromthe search engine’s perspective.

Researchers from the University of Indiana Bloomington investigated suchclaims in a fortunate series of works of some years ago [21, 40, 22] that webriefly resume. First of all, they built a mathematical model of the viciouscycle comparing two ways to access web content, one with the mediation ofsearch engines (Google in the case) and one without:

• in the surfing model - a situation where users browse content followinglinks from one site to the other - we would have ti ∼ pi ∼ ki, where k isthe number of links pointing to a website, p is the estimated pagerank ofthe website and t is amount of traffic directed to the website. Here therelation between traffic and a page’s indegree is linear: t ∼ k.

• the searching model assumes that users perform search engine queries tofind content. Deriving the relations here is a bit tricky: the authors neededto find two relationships: (i) how the PageRank translates into the rank ofa result page, and (ii) how the rank of a hit translates into the probabilitythat the user clicks on the corresponding link, thus visiting the page. Withregards to (i), the authors considered a page’s indegree equivalent to itspagerank value p and the rank r of a page inversely proportional to its p,so that a page with the largest p has average rank r w 1. With regards to


(ii) they expect the traffic to a page be a decreasing function of its rankr. t ∼ r−α. In the end

t ∼ r−α ∼ (p−β)−α = pαβ ∼ kαβ

so, in the searching model, they get t ∼ kαβ : a superlinear relationshipbetween traffic and pagerank is expected to exist.

Fig. 5.2. Relationship between average traffic and in-degree.A: The theoryprediction doesn’t match the relationship observed in real data. B: Scaling relation-ship when each page has a fixed probability h of being returned in response to aquery. (Extracted from [22])

So the theoretical prediction of the searching model was that it would haveskewed the visits towards sites with high pagerank generating a vicious circleof the rich-get-richer type, a well-known internet phenomenon [6]. However,when validating the model with real data gathered from popularity indexes4the authors discovered an opposite effect, as the relationship between indegreeand traffic was far different from expected, as shown in Fig. 5.2 A.

Contrary to our expectation, the scaling relationship is sublinear; thetraffic pattern is more egalitarian than what one would predict basedon the simple search model described above or compared with thebaseline model without search. Less traffic than expected is di-rected to highly linked sites. [21]

The factor that Flammini and collagues were failing to estimate was the top-icality of web searches.

[...] we considered the global rank of a page, computed across all pagesindexed by the search engine. However, any given query typically re-turns only a small number of pages compared with the total number

4 The authors use http://www.alexa.com as a source for traffic statistics andGoogle and Yahoo for in-degree data.

5.2 Traffic distribution in the artificial network with PageRank 45

indexed by the search engine. The size of the “hit” set and the natureof the query introduce a significant bias in the sampling process. Ifonly a small fraction of pages are returned in response to a query,their rank within the set is not representative of their global rank asinduced by PageRank

The authors, then, update their model with an assumption on the searchengine querying process. If the number of hits relevant to each query is onaverage hN where h is the relative hit set size, the relationship between trafficand indegree is altered. The result t ∼ kγ(with γ = αβ) holds true only inthe limit case of h = 1 thus, after some obscure math, the authors present arevised version of the relationship, in Fig. 5.2B.

They argue that it is the wide spectrum of human interests (and conse-quently of web searches) that drives the distribution of visits towards sitesnot necessarly high ranked, but in-topic with users requirements.

5.2 Traffic distribution in the artificial network withPageRank5

We tried to test the results of Vespignani, Flammini and Fortunato in our ownmodel, which starts from very different assumptions and doesn’t use real data,but being a complete artificial model gives us a great control over the measureswe can take and the initial conditions we set in the system. For instance, notonly can we calculate the exact amount of hits that an artifact receives andthe pagerank value of each artifact, but we can also decide to increase theamount of memes in the system, in order to show how the algorithm performswhen more “ideas” are circulating.

Let us recall the PageRank exploration strategy as implemented in Meme-To-Web. At each time step users select a meme from their {beliefs} andperform a search for it in the network of artifacts. The system will return alist of artifacts carrying that meme ordered according to its pagerank value.The user will then read an amount of artifacts equal to his readingCapacity.We tested the PageRank strategy alone and together with RandomWalk. Inthis latter case, the user performs the search like in PageRank, but will onlyread an amount readingCapacity/2 artifacts, and will start following linksfrom one of the artifacts read, reading the other readingCapacity/2 artifacts.We called this Hybrid strategy.

Our results seem consistent with the claim that search engines, and thePageRank expecially, help distributing visits to artifacts in a more egalitarian

5 In reading this section, please note that we use PageRank in italics whenreferring to the filtering algorithm as implemented in the simulation; pagerankin typewritten font when referring to the numeric “pagerank” value associated toartifacts within the simulation; and PageRank in roman when referring to thename of Google’s algorithm.


way, if compared to pure random walking the network, as described in Chapter4.

To measure how the different conditions distribute traffic among artifactswe calculated the Gini Coefficient of the pageviews in the artifact population.Developed by the Italian statistician Corrado Gini in the 1910s, the GiniCoefficient is a common measure of inequality within a distribution. It isdefined as twice the area between the 45 degree line and a Lorenz curve, wherethe Lorenz curve is a graph describing the share of total income accruing tothe poorest fraction of the population6 [28]. The index ranges 0 ≤ g ≤ 1 with0 in case of perfect equal distribution, and 1 viceversa. Table 5.1 resumesthe indices in our three test cases, with PageRank giving the less unequaldistribution of hits.

Table 5.1. Gini Coefficient of link distribution

g (nMemes=1000) g (nMemes=5000)

RandomWalk 0.94 0.96

PageRank 0.87 0.70

Hybrid 0.89 0.73

To test the extent to which the searching algorithm mitigates the rich-get-richer phenomenon we ran a simulation with 5000 memes, instead of 1000,representing in such way a wider spectrum of ideas distribuited in mindsand, consequently, in the artifacts. Interestingly enough, the Gini index ofthe RandomWalk case remains almost unmodified in both conditions, but inPageRank and Hybrid - the conditions where topicality matters the most -it decreases significantly. In the RandomWalk condition, in fact, the pathsto artifacts are only dictated by the network structure which presents, as wesaw in the previous chapter, a scale-free distribution of in-links. Therefore inRandomWalk the heavvily linked artifacts are those accessed the most, re-gardless of the memes embedded. The contrary happens in the search enginecase, where traffic is driven towards artifacts with high indegree relatively to acertain meme. This translates into a more equal distribution of visits amongartifacts when search engines are employed, because visits to artifacts are notentirely structure-dependant. The distribution of links among artifacts is ex-tremely unequal in PageRank and Hybrid also (see further), but here such

6 Suppose a society is composed of 100 income-earning households. Those 100households are arranged in ascending incomes. Lorenz Curve is constructed byplotting the cumulative share of households on the horizontal axis and the cu-mulative share of household income on the vertical axis. Every point on LorenzCurve represents one such statement that the bottom x share of households hasy share of the total income.

5.2 Traffic distribution in the artificial network with PageRank 47

inequality matters less. Fig.5.3 shows the distributions of our three test casesin the two conditions of 1000 and 5000 memes and renders explicit the depen-dence of the distributions of visits to that of memes. We added, for reference,an ideal situation of totally random navigation (grey line), where at each turnagents randomly select readingCap artifacts to consume, regardless of linksor pagerank. Neither this distribution is perfectly linear, probably because ofa strong path dependence (older artifacts are more likely to be visited), butis clearly more equitable than other conditions. This seems consistent withthe model of [21], that search engines help distributing visits among websites,instead of concentrating.

0.0 0.2 0.4 0.6 0.8 1.0

0.0

0.2

0.4

0.6

0.8

1.0

Cumulative share of artifacts

Cum

ulat

ive

shar

e of

hits

● LinkFollowingPageRankMixedRandom

s

0.0 0.2 0.4 0.6 0.8 1.0

0.0

0.2

0.4

0.6

0.8

1.0

Cumulative share of artifacts

Cum

ulat

ive

shar

e of

hits

● LinkFollowingPageRankMixedRandom

Fig. 5.3. Lorenz distribution of traffic. The line of perfect equality, the diagonal,represents a situation where every artifact has the same amount of visits. The lineof perfect inequality coincides with the horizontal and vertical axes, representing aperfectly unequal hits distribution where one artifact has all the hits and everyoneelse has none. Sandwiched between the two lines is the Lorenz curve. In (A) theLorenz distribution for a situation with 1000 memes circulating, (B) plots a conditionwith 5000 memes. PageRank and Hybrid produce a distribution which is clearlydependent from the number of memes, and less polarized than RandomWalk.

This claim holds true also if we test the relationship between traffic andpagerank, as Flammini and Vespignani do in their works. Figure 5.4 plotssuch relationship in the networks resulting from RandomWalk - that roughlymatches the surfing model of Fig. 5.2 - and Hybrid access strategies, thatequates to the searching model plus a deal of surfing, and should representa more realistic situation, thus matching better the real data. The figurerenders clear that the employment of a search engine limits the number ofhits that a page can receive at a certain PageRank level, if compared topure link following. Together with the previous finding on the distribution of


Fig. 5.4. Relationship between average traffic and PageRank in simulatednetwork obtained with logaritmic binning of pagerank values. Google’s algorithmlimits the number of hits that a page can receive at a certain pagerank level, ifcompared to pure link following.

visits (Figure 5.3) we can conclude that even in our artificial situation of peerproduction the search engine helps the traffic distribute among sites.

5.3 Effects of PageRank on the production phase:network shape.

We have seen that using PageRank to explore the network of artifacts has alevelling effect on the distribution of visits. Users will visit more sites, thusmore sites will be in their memory at publishing phase, and a wider rangeof choices will be available in the linking phase. This dynamic echoes in thelinking phase, where we see in fact a more equal distribution of inlinks amongartifacts. As Figure 5.5 shows, both the networks resulting from the exclu-sive use of PageRank and a mix of PageRank and RandomWalk do not exibita power-law distribution of inlinks. The curve is more likely to fit a lognor-mal distribution, although the fit was not properly tested. In other words, itlooks like that the mediation performed by an information filtering technol-ogy, such as PageRank, reverberates on the production of content, skewing

5.4 Memes 49

the distribution of links away from the regularity of the power law that wouldhave emerged. Reading, as detailed above, may result in a modification or thereaders beliefs and in turn - for publishing agents - can influence the creationof new artifacts, regarding the level of what memes they contain, and the cre-ation of new links. Table 5.2 updates the stats on the simulated networks. A

Table 5.2. Network properties

Nodes Edges Power law exp.

RandomWalk - simulated 26969 144576 -1.80

PageRank - simulated 25696 167923

Hybrid - simulated 25343 192020

Stanford - dataset 281904 2312497 -1.71

Political blogs - dataset 1491 19090

public dataset showing similar characteristics to the simulated ’topical’ net-works, is that of the US Political blogosphere7, a dataset retreived in 2004during the presidential campaign. Altough limited in size, with respect to ourartificial network, the political blogosphere shows a strikingly similar distri-bution of links. It is useful to note that the characteristics shown by the blogdata set are said to be frequent in many sub-graphs of the WWW, namely inclusters of sites gathered around a specific topic of interest: if links patternsare examined within specific communities, such as university or newspaperhomepages, they exibit a more uniform, less skewed distribution model [45].

5.4 Memes

Let us now move in the field of infodiversity. We saw in Section 4.2 that simplyrandom walking on the network brought to a loss of the 8% of the memes andthe diffusion of a meme in 75% of the minds. Adding PageRank to the plainrandomwalking reverted this situation to a more infodiverse condition. Withthe same initial situation of 500 users and 5000 memes, in the Hybrid caseonly a mere 3% of the memes was lost. In addition no meme managed to infectmore than the 15% of the population.

Users here will visit a more ample section of sites, but still relevant to asmall amount of memes. This is the echo chamber prolem. The echo chamberis one of the possible problems related to the high availability of niche contentpermitted by the internet. It has been documentend with regards to politicalconvintions. One can surf a number of sites, all of them holding his particular

7 Dataset retreived from http://www-personal.umich.edu/~mejn/netdata/


● ●●

●

●

●

●●

●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●

●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●

●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●

●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●

●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●

●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●

1 5 10 50 100 500

1e−

041e

−03

1e−

021e

−01

1e+

00

Index

none

.indg

● LinkFollowingPageRankMixed

● ●●

●

●

●●

●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●

●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●

●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●

●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●

1 5 10 50 100 500

1e−

041e

−03

1e−

021e

−01

1e+

00

Index

mix

.indg

● MixedPageRankPolitical blogs

Fig. 5.5. A. Distribution of inlinks in Pagerank and Hybrid. The topic-sensitive algorithms generate networks which do not follow a power law. They miti-gate the concentrating effect of the link structure and show a similar trend with thePolitical blogs network (right plot). B. Topic sensitive algorithms comparedwith blog dataset.

point of view, and never encounter a very different perspective. Pagerank, andall the topical engines vare prone to this issue

6

Collaborative filters. The wisdom of crowds

The term ’crowd’ never enjoyed a good fame in psychological and sociologicalliterature. Since Gustave Le Bon published the classic The Crowd: A Studyof the Popular Mind in 1895 [9], the concept of de-individuation projected asinister shadow over the interpretation of crowd behavior. A hundred yearslater the crowd is still a fascinating research issue, yet with an opposite con-notation, as it - now equipped with wisdom - appears to be the cornerstoneof what could represent a paradigm shift in the epistemic practices of mod-ern societies. As we have seen in Flickr and Google cases in previous chapters,wisdom of crowds systems rely on ratings expressed by users to provide mean-ingful classifications of content. The origin of this approach in online platformsdates back to the first reputation systems associated with c2c commerce, likethe eBay feedback system, and has then extended to other areas of the webnot related to commerce.

Internet Reputation systems in fact are technologies that harness thepower of one ancient social artifact of humanity - reputation/gossip - to dis-tributely regulate online communities, specifically signalling cheaters and/orgood performers, i.e. enforcing social norms in absence of a central authority.In a notable interpretation [27] social norms are an evolutionary artefact thatwould have emerged to ensure cooperation in early human societies, Conte andPaolucci [14] propose that reputation can be an evolutionary culturalartefact as well: one that ensures an easier enforcement of social norms ofcooperation, if compared to pure top-down solutions (e.g. court sanctioning),especially when implemented as an integration of them. Reputation in thissense is seen as an agent property that results from transmission of beliefsabout how this agent is evaluated with regard to a socially desirable con-duct – be it cooperation or altruism, reciprocity, or law abiding. Althoughnot deliberately designed to achieve social order, reputation-based systems infact prescribe socially acceptable conducts, like benevolence or altruism, andforbid socially unacceptable ones, like cheating or free riding. Like emergentorder reputation systems are decentralised, based upon distributed social con-trol, because each participant wants other group members to comply with the

52 6 Collaborative filters. The wisdom of crowds

group norms. Social control is known to be most effective when it is sponta-neous and distributed, this is true in natural societies, and more generally inany complex social environment where socially desirable behaviour happensto contrast with individually successful behaviour, and where legal sanctionsmay be inapplicable or costly. Or when a centralized institution does not ex-ist. No central authority exists, for example, in online auction websites, toprovide enforcement of contracts stipulated by perfect strangers, possibly lo-cated thousands of miles away. Sellers advertise goods the quality of whichthe buyers can’t verify: information asymmetry should drive the market toan adverse selection condition, turning it into a market for lemons, as Akerlof[1] described in a classical paper. That has not been the case in eBay, andthis is due to the technology of reputation that helps addressing the problemby signalling cheaters and defecting users. The earliest widespread setting tofeature an ad-hoc designed reputation technology was, in fact, the eBay feed-back forum, developed in 1996 by Pierre Omidyar [37]: a convenient systemfor users to rate business partners, crediting the honest and convenient andsanctioning the cheater. Not immune from biases and occasionally gamed [48],but yet effective in creating a whole new commercial segment and mass phe-nomenon, that of internet auctions. Wisdom of crowd - still ante litteram -was born there: a platform for the community to signal and filter out thosenot conforming with shared norms: the evolution of hive mind platforms startsright from eBay’s feedback system. Two years later, in 1998, hackers got theirown reputation system, Slashdot.org, which is probably the mosty articulatedand elegant platform for collaborative news filtering to date.

Anyone can contribute stories to Slashdot, but editors decide whatappears. Anyone is then free to comment on, and rate, the story.Where Slashdot ups the ante is that the comments themselves arerated according to whether the community thinks they are useful.Anonymous comments enter the system rated at zero points; com-ments signed by new (untrusted) names enter at one point; commentscoming from veterans who have demonstrated their authority in thepast start at two. Once a comment is made, other previous contrib-utors are randomly selected to assess whether the comment’s ratingshould be raised or lowered. Contributors are allocated five points,with which they can raise or lower ratings by one point only, butcannot moderate posts which they have chosen to comment on. Slash-dotters who contribute quality stories are rewarded with ’karma’, atime limite dpoint system allowing them to increase their reputationcapital. These moderation points expire after five days, so they cannotbe accumulated. [42]

Slashdot’s metamoderation system is the first implementation of a reputa-tion architecture for filtering content and attributing authority. All the subse-quent systems, notably social content sites like reddit.com, digg.com drew onthe principles set by Slashdot. Social news sites are a typical offspring of the

6 Collaborative filters. The wisdom of crowds 53

Web2.0 approach, they can be described as man-made, self-organizing, com-plex informational systems that filter out noise from the signal. Here usersselect and discuss upon "interesting" and valuable news headlines proposedthrough a process of collaborative content filtering and displayed as a rankedlist on the homepage. The process of selection is straightforward and typicalof the web2.0 "smartmobs" approach: users post links to news items discov-ered on the web; fellow users can comment each post and vote it "UP" it ifthey consider it interesting, relevant, or generally "worth reading", or downit if not useful/interesting, inconsistent. The reputation mechanism of thesesites is simpler than that of Slashdot: here users submitting popular (i.e. thosethat get many "ups") stories or valuable comments gain in ’karma’, or ’repu-tation points’. Posts with a certain amount of positive votes collected over acertain amount of time - and sometimes submitted by "reputable" users - getfeatured in the homepage. The same applies to the debate that is triggeredaround single pieces of content: users rate others’ comments and the mostwell received appear on the top of the list, while those falling below a certainthreshold become invisible. The voice of high-karma users here doesn’t countmore towards the reaching of the threshold needed for a post to get on thefrontpage, or a comment to be displayed. The most well known social newsplatforms, Reddit.com and Digg.com, have been growing exponentially in theiruser base. With several millions of users1 collectively filtering and discussingnews items, they started presenting themselves as grassroot challengers toprofessional mainstream media newsdesks in the gatekeeping process. At thesame time heavy criticism was raised and the alleged democratic nature ofthese media was questioned, as it became clear that only a small amount of"power users" was responsible for the most part of the frontpage of Digg.com2.If we consider Digg as archetypical of the whole class of tools discussed here,the "democracy" problems can be seen as a canary value of the huge biasesthat probably, unquestioned, affect these platforms.

From an implementation point of view this class of systems poses a seriousquestion, that of the nature of evaluation carried on by users. If in the eBaycase the social norm enforced by the reputation system, and with regards towhich the evaluation is performed, is clear and recognized by all the partici-pants, it is not the case for Reddit and Digg. What is evaluated in eBay is theperformance of a business partner in delivering what he promises - quality ofthe goods sold or promptness in the payment - quite ’objective’ parameters.A social norm of fairness, one could say, is being distributely enforced by eco-nomic reputation systems. In Slashdot no economic transactions take place:assessing the quality of a comment or of an article is increasingly subjective.What is relevant to me could not be to somebody else; what is funny to onecould be absolutely lame to somebody else. Even the parameters are fuzzy:

1 http://www.ignitesocialmedia.com/2008-social-network-analysis-report/2 http://socializingdigg.wordpress.com/2008/09/24/diggs-new-biz-model-ban-top-users-and-hit-300m/

54 6 Collaborative filters. The wisdom of crowds

what do users exactly evaluate in content filtering websites? While the Slash-dot metamoderation system has an explicit set of references - users are askedto rate articles in a scale 1-5 with regards to interestingness, fun, insightful-ness, informativeness - this does not hold for later platforms. Reddit and Diggusers do not have any reference to keep in mind when they evaluate an item.The evaluation is liquid and impressionist. Could be relevance, could be style,could be anything.

So we were posed with a major design choice to make, how to model user’sactivity in collaborative content filters for our abstract implementation. At theend we choose to design the collaborative content filter algorithm sticking tothe similarity constraint. The more memes contained in an artifact are alsoheld by the reader, the more likely she will be to upvote the artifact. Formally

pV ∼∑

[m ⊂ {memes} ∧m ⊂ {artmemes}]

In addition, we don’t add reputational mechanisms like in Slashdot, thevote of each user counts as that of anyone else, in the more ’democratic’fashion of, for example, Reddit.

When employing this content access strategy users will be presented a listof the most upvoted artifacts, among which they will chose readingCapacityartifacts to read. The test we performed on this condition consisted in addingit to the Hybrid situation: at each turn one third of readingCapacity artifactsis discovered through the PageRank system, another third is taken from theartifacts with the most upvotes and artifact links are followed reading theremaining third of readingCapacity.

Infodiversity: preliminary results

Adding a collaborative content filter to our simulation heavily changed thefigures in the memetic distribution: with the condition used in Sec. 4.2 and5.4 the addition of a collaborative filter didn’t change much the total num-ber of memes circulating in the system, the figures are comparable to thoseoccurring in the PageRank case, but favours the diffusion of memes. UsingReddit in fact let a meme spread in the totality of the population, while onemore managed to reach 90% af agents’ minds. Table 6.1 presents a roughcomparison between the three experimental conditions of RandomWalk, Ran-domWalk+PageRank and RandomWalk+PageRank+Reddit. What emerges isthat the evolution of the web content filtering platforms began with a veryconcentrational technology, the hyperlink system itself, which was followed byan extraordinary levelling system, the PageRank, which lets traffic distributemore equally among artifacts, but renders the discovery of new content adifficult task. Collaborative content filters, finally, let very little informationextinguish from the environment, but at the same time allow for a certaindeal of information to spread fast and take over the entire society.

6 Collaborative filters. The wisdom of crowds 55

Table 6.1. Memetic diversity in the three experimental conditions

1% 25% 50% 75% 100% Total count % variation

RandomWalk 23% 1 1 1 0 4418/4800 -8

RW+PageRank 33% 0 0 0 0 4648/4802 -3.3

RW+PR+Reddit 38% 8 6 2 1 4657/4798 -2.9

7

Discussion and (hopefully) future work

In this work we tried to paint a possible road for an agent-based investigationof internet based mechanisms of aggregation and their outcome on culturaldiversity in the society. The aggregation platforms, organized according tothe principle of wisdom of crowds, play a huge epistemic function, in thatthey shape the access to the collective memory and the cultural production ofsociety. Such function is made even more critical by the emergence of a newparadigm in the production of cultural goods, founded on the universal avail-ability of the main means of production, the absence of reproduction costs andthe status of common of the first matter. This new paradigm is set to increasethe production of cultural artifacts especially by non professional publishersand redefining, at least in the field of symbolic/cultural goods, the relationsof production, introducing new economic actors and relying on completelydifferent premises, if compared to the traditional cultural industry model. Wereviewed the work of Yochai Benkler, who more than anyone else tried tofigure out the possible economic (but also political and social) implicationsof the phenomenon, from an enthusiastic point of view. We then examined acritical position towards relying on automated and unscrutinized technologiesfor performing complex social functions related to the organization of culture.We have advanced an interpretation of the nature of web based epistemictechnologies, locating their origins in the early hacker culture, in the culturalclimate of the 1970s in the USA and in particular in the notion of democracyand charismatic leadership that it carries.

In the core of this work we have advanced a possible method of anal-ysis, agent based simulation, and proposed a computational model, calledMeme-To-Web, for exploring the effect of epistemic algorithms upon informa-tion diversity in infosocieties. The model simulates a society based on peer-production of cultural goods and technology mediation in access to culturalproduction, with a set of hypotheses on human behaviour in online contextsderived from literature and observation. The proposed model employs the con-troversial metaphore of the ’meme’ to represent an atomic unit of knowledgeembedded at the same time in human minds and in the artifacts that spring

58 7 Discussion and (hopefully) future work

from such minds’ activity. We have tried to keep the model ’rooted in reality’validating its outcomes by comparing them to those of a fair amount of recentrelevant research and datasets, and found the model to produce reasonableoutputs - namely network shapes that appear to match those commonly ob-served, and browsing patterns that reproduce those documented with differenttheoretical instruments.

We have also advanced a rough analogy between the notion of biodi-versity, that is the amount of different species that an ecosystem hosts andinfodiversity, the amount of different ’ideas’ (in our terms, memes) that aninfosystem allows to circulate.

Some preliminary conclusions have been drawn: the evolution of wisdomof crowds architectures, in the last decade, seems to oscillate between con-centration and “inclusivity” in a fairly regular fashion, and the system - theinternet - seems to have evolved towards a sort of equilibrium, where ideasare almost all represented, but only a few can spread extremely widely.

To a strongly concentrational technology - the hyperlink structure of theweb itself, organizing artifacts in networks that favor concentration, character-ized by the scale free principle - followed a formidable equalizing technology,Google’s PageRank, that helps distributing access towards a number of prod-ucts, matching the richness of the environment but produces a sort of ’echochamber’, in that one finds interesting content, but a system with only amechanism for surfing links and one for finding keyword-based content leaveslittle chance for ideas to spread widely. Then came collective filters and rat-ing systems, that seem to have a balancing effect on the system. Adding thismechanism of selection - the only one in which every member of the commu-nity can potentially have a say - resulted in very few memes to disappear,and at the same time a few memes spreading to the extent of infecting all thepopulation.

However, it is no mystery that these are very partial and preliminary re-sults and that the research presented here doesn’t live up to all the promisesformulated in Chapters 1 and 2. This is mostly due to time and intellec-tual limitations of the author: for instance, an appropriate analysis of thememetic distribution in the society accounting for thematic subgroups, emer-gent memeplexes, and other phenomena surely present in the model’s outputsand in need of investigation, would have required a much higher mastery ofstatistics and mathematical methods.

Moreover, one key assumption of the implementation presented here, thatof a random initial distribution of memes, probably needs to be dropped, or atleast re-thought. We should give an initial distribution of memes in order totest the performance of content filters over, for example, polarized societies.The initial randomly distribution of memes is a limit case, which should beused, at best, as a control condition, but to test the outcomes of contentfiltering in a more realistic context it needs to be changed. Users are not bornin peer production systems, they enter with their ideas already distributedin extremely unequal ways. It would be extremely interesting to investigate

7 Discussion and (hopefully) future work 59

the outcomes of each system over heavily polarized societies, such as thosewe live in, with their taboos, their dangerous ideas, and their common senses.Different kinds of society could be implemented: for example one with a smallamount of widely diffused memes, forming the core beliefs of the group, plusa bigger amount of memes only present in certain sectors of the population.Today’s netizens come from cultures heavily influenced by broadcasting media,they bring an extremely skewed distribution of memes. The model should thenbe employed to test the performance of systems in presence of phenomena suchas the neverending september, as described in section 2.3. This means thatnew users should not be entering the system with a regular constant flow, butmany at a time, with a particular memetic configuration.

To sum up, we believe that the meme-to-web is scalable and flexible enoughto embed these changes with very little effort and the model presented and theapproach suggested are valid and paint the road for a series of work that couldbring more light over this profound transformation that society is undergoing.

References

1. George A. Akerlof. The market for "lemons": Quality uncertainty and the mar-ket mechanism. The Quarterly Journal of Economics, 84(3):488–500, 1970.

2. Chris Anderson. The Long Tail: Why the Future of Business Is Selling Less ofMore. Hyperion, July 2006.

3. Robert Axelrod. The dissemination of culture: A model with local convergenceand global polarization. Journal of Conflict Resolution, 41(2):203–226, April1997.

4. A. L. Barabasi and R. Albert. Emergence of scaling in random networks. Sci-ence, 286(5439):509–512, October 1999.

5. Albert-Laszlo Barabasi. Linked. Perseus Publishing, 2002.6. Albert-Laszlo Barabasi. Linked: How Everything Is Connected to Everything

Else and What It Means. Plume, reissue edition, April 2003.7. J.P. Barlow. Postcards from the net: An intrepid guide to the wired world, 1996.8. Yochai Benkler. The Wealth of Networks: How Social Production Transforms

Markets and Freedom. Yale University Press, October 2007.9. Gustave L. Bon. The Crowd: A Study of the Popular Mind. Classic Books

Library, March 1896.10. S. Brin and L. Page. The anatomy of a large-scale hypertextual web search

engine. Computer Networks and ISDN Systems, 30(1-7):107–117, April 1998.11. Cristiano Castelfranchi. Towards a cognitive memetics: Socio-cognitive mech-

anisms for memes selection and spreading. Journal of Memetics, 5(1), March2001.

12. Manuel Castells. The Internet Galaxy : Reflections on the Internet, Business,and Society. Oxford University Press, April 2001.

13. Rosaria Conte and Cristiano Castelfranchi. Cognitive and social action. UCLPress, London, 1995.

14. Rosaria Conte and Mario Paolucci. Reputation in Artificial Societies: SocialBeliefs for Social Order. Springer, October 2002.

15. Richard Dawkins. The Selfish Gene. Oxford University Press, USA, 3 edition,May 2006.

16. Terrence Deacon. The trouble with memes and what to do about it. TheSemiotic Review of Books, 10(3), 1998.

17. Bruce Edmonds. The revealed poverty of the gene-meme analogy. why memeticsper se has failed to produce substantive results. Journal of Memetics - Evolu-tionary Models of Information Transmission, 9, 2005.

62 References

18. J. M. Epstein. Agent-based computational models and generative social science.Complexity, 4(5):41–60, 1999.

19. Christos Faloutsos, Mary McGlohon, Jure Leskovec, and Micaela Götz. Model-ing blog dynamics. In AAAI Conference on Weblogs and Social Media, 2009.

20. Carlo Formenti. Cybersoviet. Utopie postdemocratiche e nuovi media. RaffaelloCortina Editore, 2008.

21. S Fortunato, a Flammini, F Menczer, and a Vespignani. Topical interests andthe mitigation of search engine bias. Proceedings of the National Academy ofSciences of the United States of America, 103(34):12684–9, August 2006.

22. Santo Fortunato, Alessandro Flammini, Filippo Menczer, and AlessandroVespignani. The egalitarian effect of search engines, Nov 2005.

23. G. Gigerenzer, U. Hoffrage, and H. Kleinbölting. Probabilistic mental models: abrunswikian theory of confidence. Psychological review, 98(4):506–528, October1991.

24. Nigel Gilbert. Agent-Based Models (Quantitative Applications in the Social Sci-ences). Sage Publications, Inc, annotated edition edition, September 2007.

25. Nigel Gilbert and Klaus G Troitzsch. Simulation for the Social Scientist. OpenUniversity Press, 2005.

26. Jim Giles. Internet encyclopaedias go head to head. Nature, 438(7070):900–901,December 2005.

27. G. R. Grice. The emergence of norms by edna ullmann-margalit clarendon press:Oxford university press, 1977, xiii + 206 pp., Â£8.00. Philosophy, 54(209):420–421, 1979.

28. Mark S. Handcock and Martina Morris. Relative Distribution Methods in theSocial Sciences (Statistics for Social Science and Behavorial Sciences). Springer,August 1999.

29. Pekka Himanen. The Hacker Ethic. Random House Trade Paperbacks, February2002.

30. Myshkin Ingawale, Rahul Roy, and Priya Seetharaman. Persistence of culturalnorms in online communities: The curious case of wikilove. In Pacific AsiaConference on Information Systems, 2009.

31. Amit Karandikar, Akshay Java, Anupam Joshi, Tim Finin, Yelena Yesha, andY. Second space: a generative model for the blogosphere. ICWSM. AAAI.,2008.

32. Charles J. Krebs. Ecological Methodology (2nd Edition). Benjamin Cummings,July 1998.

33. Kalevi Kull. Copy versus translate, meme versus. European Journal for SemioticStudies, (1):101–120.

34. Jure Leskovec, Kevin J. Lang, Anirban Dasgupta, and Michael W. Mahoney.Community structure in large networks: Natural cluster sizes and the absenceof large well-defined clusters. CoRR, abs/0810.1355, 2008.

35. Pierre Levy. Collective Intelligence: Mankind’s Emerging World in Cyberspace.Perseus Books Group.

36. Steven Levy. Hackers: Heroes of the Computer Revolution. Penguin Putnam,January 2002.

37. Lingfang I. Li. Reputation, trust, & rebates: How online markets can improvetheir feedback mechanisms. Technical report, Institute for Mathematical Be-havioral Sciences, 2006.

38. Andrew Lih. The Wikipedia Revolution: How a Bunch of Nobodies Created theWorld’s Greatest Encyclopedia. Hyperion, 2009.

References 63

39. MR Meiss, Bruno Gonçalves, JJ Ramasco, and A Flammini. Agents, Bookmarksand Clicks: A topical model of Web navigation. arxiv.org.

40. F. Menczer, S. Fortunato, A. Flammini, and A. Vespignani. Googlearchy orgooglocracy? IEEE Spectrum, 2006.

41. M. E. J. Newman. Assortative mixing in networks. Physical Review Letters,89(20):208701+, Oct 2002.

42. Mathieu O’Neil. Cyberchiefs : autonomy and authority in online tribes. PlutoPress, 2009.

43. Gloria Origgi. Designing wisdom through the web. the passion of ranking. InCollective Wisdom. Cambridge University Press, 2009.

44. M Paolucci, S Picascia, and S Marmo. Electronic reputation systems. IGI,Hershey, USA, 2010.

45. D M Pennock, G W Flake, S Lawrence, E J Glover, and C L Giles. Winnersdon’t take all: Characterizing the competition for links on the web. Proc NatlAcad Sci U S A, 99(8):5207–5211, Apr 2002.

46. Stefano Picascia and Mario Paolucci. Cultural circulation and the pagerankeffect. In Presented at the 3rd World Congress of Social Simulation. Universityof Kassel, 09 2010.

47. Eric Raymond. The cathedral and the bazaar. Knowledge, Technology andPolicy, 12:23–49, 1999. 10.1007/s12130-999-1026-0.

48. Paul Resnick, Ko Kuwabara, Richard Zeckhauser, and Eric Friedman. Reputa-tion systems. Commun. ACM, 43(12):45–48, December 2000.

49. Roberts, Ben. Beyond the ’Networked Public Sphere’: Politics, Participationand Technics in Web 2.0. Fibreculture, (14), 2009.

50. Mirko Tobias Schafer. Bastard culture! user participation and the extension ofcultural industries, 2008.

51. C. E. Shannon and W. Weaver. The mathematical theory of communication.University of Illinois Press, 1949.

52. Ralf D. Sommerfeld, Hans-JÃŒrgen Krambeck, and Manfred Milinski. Multiplegossip statements and their effect on reputation and trustworthiness. 2008.

53. F. Stonedahl and U. Wilensky. Netlogo pagerank model, 2009.54. James Surowiecki. The Wisdom of Crowds. Anchor, August 2005.55. Klaus G. Troitzsch. Simulating collaborative writing: Software agents produce

a wikipedia. In ESSA 2008 - The Fifth Conference of the European SocialSimulation Association, 2008.

56. Peter van den Besselaar and Dennis Beckers. The life and death of the greatamsterdam digital city. pages 66–96. 2005.

57. U. Wilensky. Netlogo, 1999.

memes, networks and artifacts. cultural circulation in an artificial peer-production society

Documents