||Department of Humanities, Social and Political Sciences
Program in Computational Social Science
Social Modelling, Agent-Based
Simulation and Collective Intelligence(Week 7)
02.04.2016 1
ETH D-GESS: 851-0585-37L
Ovi Chris Rouly, PhD
||Department of Humanities, Social and Political Sciences
Program in Computational Social Science
ETH D-GESS: 851-0585-37L Week 7
02.04.2016Ovi Chris Rouly, PhD 2
Social Networks
||Department of Humanities, Social and Political Sciences
Program in Computational Social Science02.04.2016Ovi Chris Rouly, PhD 3
This lesson considers social networks as abstract, mathematically
tractable, and computationally instantiable systems. We will look
at ways to model, instantiate, manipulate, and to analyze them.
Within the fields of computational social systems modeling, the
applied “deep-study” of Social Networks has become a discipline
unto itself. Moreover, its paradigms are based upon broadly
dissimilar scientific footings. To examine even a few of those
foundation stones we will have to consider mathematics (graph
theory), computer science (algorithms), sociology (population
group trends), psychology (individual and social behavior), and
complex networks.
Let’s get started!
||Department of Humanities, Social and Political Sciences
Program in Computational Social Science02.04.2016Ovi Chris Rouly, PhD 4
Social Modelling, Agent-Based
Simulation and Collective Intelligence
Course Overview
Procedure (Parts I & II):
1. Examine a selection of published, formal models of social processes
2. Learn how to analyze and extend simple models and to develop your own social process models using existing computer-coded examples
||Department of Humanities, Social and Political Sciences
Program in Computational Social Science02.04.2016Ovi Chris Rouly, PhD
Derived from “Introduction to Computational Social Science”, Cioffi-Revilla, 2014.
A Social Network as a Descriptor of Social Trends1. Nodes are elements within an abstract set
a. Individual object actors (animate, inanimate)
b. Ideas, symbols, groups-of-nodes, entire networks
2. Edges are possibly directed, always explicit, tangible/intangible relationships
a. Weighted bindingsb. Semantic bindings
c. Information-conduits
3. Aggregations are pregnant with implicit social meanings
a. Dyads
b. Triadsc. Cliques
4. Descriptive properties over a network include
a. Within and between element and network aggregations
b. Betweenness, centrality, and connectedness measures
c. Span, density, embeddedness, degreed. Simmelian ties, In-group/Out-group, and many others
d. “Clans”
e. Clusters
d. implicitly temporal
||Department of Humanities, Social and Political Sciences
Program in Computational Social Science02.04.2016Ovi Chris Rouly, PhD
An abstract topology with implications for algorithm design and social modeling
A Network as an Abstract Topology and Social “Glue”
1. An agent can be associated with (mapped onto) a network node
(Here the agent moves only between nodes)a. An agent can move only if there is a connecting edge
b. An agent can move regardless of edge connections
2. A network node can be associated with (mapped onto) an agent
(Here the agent can move anywhere in raster, vector, or logical space)
a. An agent can move only if there is a connecting edge
b. An agent can move regardless of edge connections
||Department of Humanities, Social and Political Sciences
Program in Computational Social Science02.04.2016Ovi Chris Rouly, PhD
From “Information Theory of Complex Networks”, Solé & Valverde, 2004, p. 4.
Regularity versus Complexity
ER = Erdos- Reyni
SF = Scale-Free
||Department of Humanities, Social and Political Sciences
Program in Computational Social Science02.04.2016Ovi Chris Rouly, PhD
Networks: Abstract entities that are artifacts of social interaction
Relative Size versus Relative Complexity
||Department of Humanities, Social and Political Sciences
Program in Computational Social Science02.04.2016Ovi Chris Rouly, PhD
“Error and attack tolerance of complex networks”, Albert, Jeong, Barabási, 2000.
Network Complexity versus Connectivity
(An Example: Exponential Network Compared to Scale-free Network)
Exponential(5 most connected nodes reach 27% of others)
Scale-free(5 most connected nodes reach > 60% of others)
||Department of Humanities, Social and Political Sciences
Program in Computational Social Science02.04.2016Ovi Chris Rouly, PhD
Taken from “Towards Emergent Social Complexity”, Rouly, 2015, p. 124.
A Large, Regular, Artificial “Genealogical” Network
approx. 2,400 years
||Department of Humanities, Social and Political Sciences
Program in Computational Social Science02.04.2016Ovi Chris Rouly, PhD
“How robust is the Internet?”, Tu, 2000.
A Large, Scale-Free, Human-made Network
||Department of Humanities, Social and Political Sciences
Program in Computational Social Science
G = { N, E, D }
where
• D is set of dimensions, {d1, d2, ... d|D|}
and each d is a triple over G
where
• dk ∈ D; 1 < k ≤ |D|
G = { N, E }
where
• G is a graph
• N is a set of nodes, {n1, n2, ... n|N|}
• E is a set of edges, {e1, e2, ... e|E|}
and an edge is a tuple over G
where ni, nj ∈ N; i ≠ j; i, 1 < j ≤ |N|
02.04.2016Ovi Chris Rouly, PhD
Quantitative operations over a network matrix can amplify social insight.
Social Networks as Multidimensional Graphs
||Department of Humanities, Social and Political Sciences
Program in Computational Social Science02.04.2016Ovi Chris Rouly, PhD 13
Manipulation of the Information in a Social Network
Clearly, on a large network, the process of transcribing network
data from a “relationship graph” onto a matrix could be tedious.
But, once mapped onto a matrix, we can manipulate the network
algebraically and analyze the underlying system.
If we take this process one step further (automating the
transcription and putting everything in a “machine readable” form)
then, we can begin to do Social Network Analysis (SNA).
http://vlado.fmf.uni-lj.si/pub/networks/pajek/doc/pajekman.pdf
||Department of Humanities, Social and Political Sciences
Program in Computational Social Science02.04.2016Ovi Chris Rouly, PhD 14
Measures exist to quantify properties of networks; some more useful than others.
Social Network Analysis – a self-describing process whose
intention is to identify meaningful features in a related set of artefacts
associated with social beings.
1. There exists no single “best measure” to describes a node (or an edge)
as most important in a network.
2. Measures exist to identify Bridging or Spanning nodes
3. Cliques and Clusters of nodes can be found and relative node Density
4. Measures of relative Connectedness and Centrality are numerous
For example:
a. Degree Centrality – how many edges attach to the node
b. Closeness Centrality – a node’s relative closeness to other
nodesc. Eigen (vector) Centrality – importance of a node influences the
importance of other nodes connected to it
d. Betweenness Centrality – situated placement in the network
||Department of Humanities, Social and Political Sciences
Program in Computational Social Science02.04.2016Ovi Chris Rouly, PhD 15
That node whose central betweenness (CB) is greatest represents
among all possible paths between all other nodes, over all nodes in a
network, the most important. This is the “queen bee” of the network.
"... a point in a communication network is central to the extent that it falls on the shortest path between [all] pairs of other points" (Freeman, 1977).
Opinion:
While many network measures exist, central betweenness (CB) “may” be
the most important to us for understanding networks that seek to illustrate
the communication of ideas, substance, and information in general.
CB 𝑖 =
𝑗<𝑘
𝑔𝑗𝑘 (𝑖)
𝑔𝑗𝑘, 𝑖 ≠ 𝑗 ≠ 𝑘
X
||Department of Humanities, Social and Political Sciences
Program in Computational Social Science02.04.2016Ovi Chris Rouly, PhD 16
Small World Networks are defined as those networks where the
distribution of the lengths (L) of all possible paths between all possible
nodes (N) in the network can be approximated by a log relation over the
number of nodes in the network. In particular, a Small-World Network will
have a relatively high coefficient of clustering (Ci) and a set of relatively low overall path lengths (e) over all nodes (i).
Opinion:
Among the naturally occurring network configurations, the Small-World
Network may be among the most important. It is a network type whose
interconnection configuration often emerges as an artifact associable with living systems. Watts & Strogatz (1998) pointed out they could be found in
structures as diverse as the, “neural network of the worm Caenorhabditis
elegans, the power grid of the western United States, and the collaboration
graph of film actors [movie stars] ...”.
L α log N
||Department of Humanities, Social and Political Sciences
Program in Computational Social Science02.04.2016Ovi Chris Rouly, PhD 17
Clockwise from upper left:
LinkedIn network of DJ Patil (former chief scientist at LinkedIn)
A sexual network associated with an STD outbreak. Sexually Transmitted Infections,
Potterat et al. doi:10.1136/sti.78.suppl_1.i152
A social network in which clustering of body weight is visible; node size corresponds to
body weight, yellow nodes depict a BMI ≥30 (obese). The New England
Journal of Medicine, Christakis & Fowler doi:10.1056/NEJMsa066082
Taken from https://blogs.unimelb.edu.au/sciencecommunication/2012/09/25/its-a-small-world-after-all/
||Department of Humanities, Social and Political Sciences
Program in Computational Social Science02.04.2016Ovi Chris Rouly, PhD 18
a spatial agent-based
model. And, we will talk
about how we could
integrate an agent-
based simulation (like
Schelling’s) together
with an abstract social
network.
Of course, we will also
propose a reading
assignment and discuss
the class deliverables.
After a break we will look at few example networks. We will also
discover that we now have many of the basic tools needed to build an
spatial agent-based model that incorporates a network-topology onto
||Department of Humanities, Social and Political Sciences
Program in Computational Social Science
5-6 minutes
02.04.2016Ovi Chris Rouly, PhD 19
break
||Department of Humanities, Social and Political Sciences
Program in Computational Social Science
Albert Einstein
02.04.2016Ovi Chris Rouly, PhD 20
"Things should be made as simple as possible - but no simpler."
||Department of Humanities, Social and Political Sciences
Program in Computational Social Science02.04.2016Ovi Chris Rouly, PhD 21
Brief Primer: Social Psychology and Social Network Analysis
• Nodes – are the primary elements in a network. Their function in the network
is variously either “self” (ego) or “other” (alter). If “self” then, their purpose is
“self-motivated” action relative to their role and their subjective network
knowledge. If “others” their function is that of an arbiter or reactive agent.
• Edges – represent social connectivity in a network. They represent evidence of physical, informational, and or some other material or nonmaterial transfer
or contact between nodes. Typically edges suggest some social binding
between individuals and or groups of nodes. Finally, an edge often connotes
implicit temporal properties.
• Dyads – this is a two node network construct. Graphed together with an edgethey can show flow direction or not. Examples include:
• Triads – these closed network constructs have three nodes.
• Simmelian ties – (named for Georg Simmel) and extended by Krackhardt
(1999), are strong, bidirectional social bindings. Simmel thought that within a
triad having symmetrical Simmelian ties, the “self” would not be lost (Simmel & Wolff, 1950).
• Clique – a sub-network of three or more nodes having similar social purpose.
||Department of Humanities, Social and Political Sciences
Program in Computational Social Science
Network analysis tools reveal features often obscure but sometimes useful
02.04.2016Ovi Chris Rouly, PhD 22
Concepts:
Modeling paradigm – social networks
Analytic Tool – Pajek (others are available UCINet, RedSqirl, Gelphi, Cytoscape ...)
Process hypothesis – simple SNA can reveal hidden data insights
Networks: { Employees of: “IBM” &
“Google,” Customers of “Credit Suisse” &
“UBS,” and an amateur bicycle team }
Results*:
Centrality = 0.16363636
Degree = 5 (max)
Betweenness = 0.12644628
Most Connected = Harry & Credit SuisseMost Between = Credit Suisse* over all nodes
Analysis of our simple pedagogical social network
||Department of Humanities, Social and Political Sciences
Program in Computational Social Science02.04.2016Ovi Chris Rouly, PhD 23
Concepts:
Modeling paradigm – multi-dimensional social network analysis
Analytic Tool – VOSViewer (others are available CitNetExplorer, CiteSpace, Histite ...)
Mechanism hypothesis – multi-dimensional SNA can reveal yet more data insights
Published in the Journal of the American
Society for Information Science and Technology
“... In our experimental analysis, we compare
three approaches for constructing bibliometric
maps... (Van Eck et al, 2010, p. 2).
In this example an analysis of the social
network data describes relationships between
authors, journals and keywords.
A comparison of two techniques for bibliometric mapping:
Multidimensional scaling and VOS (Van Eck, N., Waltman, L., Dekker, R.
& Van den Berg, J., 2010)
||Department of Humanities, Social and Political Sciences
Program in Computational Social Science02.04.2016Ovi Chris Rouly, PhD 24
Authors
Journals
Keywords
“... In our experimental analysis, we compare three
approaches for constructing bibliometric maps...
(Van Eck et al, 2010, p. 2).
In this example an analysis of the social network
data describes relationships between authors,
journals and keywords.http://www.vosviewer.com/download
||Department of Humanities, Social and Political Sciences
Program in Computational Social Science02.04.2016Ovi Chris Rouly, PhD 25
Concepts:
Modeling paradigm – complex networks
Process hypothesis – implicit system data exists in human social activity
Published in PLoS One
“A plausible approach to assess the
impact of a natural, large-scale disruption
is to measure systematic changes in the
distributional form of these standard
centrality measures. However, we find
that the functional form of degree, flux,
and betweenness distributions is
surprisingly robust to these disruptions as
illustrated in Fig. 2.” (Woolley-Meza et al,
2013, p. 3).
Eyjafjallajokull and 9/11: The Impact of Large-Scale Disasters on
Worldwide Mobility (Woolley-Meza, O. Grady D., Thiemann C., Bagrow, J. &
Brockmann, D., 2013)
“Figure 2. Network properties before and after
a natural disruption” (p.3). [P(b) is a measure of betweenness probability before disruption.]
||Department of Humanities, Social and Political Sciences
Program in Computational Social Science
Week 7 deliverable: Read then write to persuade or dissuade for SNA.
02.04.2016Ovi Chris Rouly, PhD 26
Reading assignments:
A chapter, if you want a know a little:
Scott, J. (1988). Social network analysis. Sociology, 22(1), pp. 109-127.
A book, if you want to know a lot more:
Scott, J. (2012). Social network analysis. Sage.
Writing assignment:
Write a 1-2 page White Paper arguing for (or against) the face validity of
using Social Network Analysis to describe the socio-political operations of
one of the individual country governments within the European Union. You will need at least two SNA citation/references to support your argument.
Nota bene: Any explicit reference to people or organizations *MUST* be
cited to, referenced by, no less than 2 other sources. Failure to adhere to
this requirement will result in a reduction of grade.
Deliverable in two weeks
||Department of Humanities, Social and Political Sciences
Program in Computational Social Science02.04.2016Ovi Chris Rouly, PhD 27
• Albert, R., Jeong, H., & Barabási, A. L. (2000). Error and attack tolerance of complex networks.
Nature, 406(6794), pp. 378-382.
• Cioffi-Revilla, C. (2014). Introduction to computational social science: principles and applications.
Springer Science & Business Media. Chapter 4.
• Freeman, L. C. (1977). A set of measures of centrality based on betweenness. Sociometry, 35-41.
• http://faculty.ucr.edu/~hanneman/nettext/
• http://mrvar.fdv.uni-lj.si/pajek/
• http://vlado.fmf.uni-lj.si/pub/networks/pajek/doc/pajekman.pdf
• Krackhardt, D. (1999). The ties that torture: Simmelian tie analysis in organizations. Research in the
Sociology of Organizations, 16(1), pp.183-210.
• Batagelj, V., & Mrvar, A. Pajek–Program for Large Network Analysis. Home page: http://mrvar.fdv.uni-
lj.si/pajek/. (accessed on 11 March, 2016)
• Solé, R. & Valverde, S. (2004). Information theory of complex networks: on evolution and
architectural constraints. In Complex networks. Springer Berlin Heidelberg. pp. 189-207.
• Simmel, G. & Wolff, K. (1950). The sociology of Georg Simmel (Vol. 92892). Simon and Schuster.
• Tu, Y. (2000). How robust is the Internet?. Nature, 406(6794), pp. 353-354.
• Van Eck, N., Waltman, L., Dekker, R. & Van den Berg, J. (2010). A comparison of two techniques for
bibliometric mapping: Multidimensional scaling and VOS. Journal of the American Society for
Information Science and Technology, 61(12), 2. pp. 405-2416.
• van Eck, N. J., & Waltman, L. (2014). Visualizing bibliometric networks. (Chapter 13). In Measuring
scholarly impact. Springer International Publishing. pp. 285-320.
REFERENCES
||Department of Humanities, Social and Political Sciences
Program in Computational Social Science02.04.2016Ovi Chris Rouly, PhD 28
• Watts, D. J. & Strogatz, S. H. (1998). Collective dynamics of 'small-world' networks. Nature 393
(6684). pp. 440–442.
• Woolley-Meza, O., Grady, D., Thiemann, C., Bagrow, J. P., & Brockmann, D. (2013). Eyjafjallajökull
and 9/11: the impact of large-scale disasters on worldwide mobility. PLoS one, 8(8), e69829.
REFERENCES
||Department of Humanities, Social and Political Sciences
Program in Computational Social Science02.04.2016Ovi Chris Rouly, PhD 29
consider theories about how opinions are formed and propagate
crowd disasters and ways to mitigate them
pedestrian traffic as emergent social phenomena
and, discuss Collective Intelligence and can it be instantiated
In the weeks that follow we will:
||Department of Humanities, Social and Political Sciences
Program in Computational Social Science
ETH Zurich
D-GESS Computational Social Science
Clausiusstrasse 50
8006 Zürich, Switzerland
http://www.coss.ethz.ch/
Ovi Chris Rouly, PhD.
Email: [email protected]
Telephone: (41) 044-633-8380
© ETH Zurich, 2 April 2016
02.04.2016Ovi Chris Rouly, PhD 30
Contact information