A Framework for Recommender Systems in Online Social NetworkRecruiting: An Interdisciplinary Call to Arms

Ricardo BuettnerFOM University of Applied Sciences, Institute of Management & Information Systems, Germany

ricardo.buettner@fom.de

Abstract

I sketch an interdisciplinary framework for recom-mender systems searching online social networks forfuture employees. In contrast to previous approachesmy framework covers the whole person-organizationenvironment (P-OE) fit, and so it also includes crucialsocial components (e.g. workgroup fabric, and orga-nizational culture). In a rudimentary way I show howinformation extracted from online social networks canbe used to determine the P-OE fit. Due to a moreaccurate way of calculating the P-OE fit my frameworkfacilitates higher quality recommendations. Further-more my framework enables a better understanding ofthe P-OE fit problem.

1. Introduction

Human Resource Management (HRM) has seen arecent influx of electronic support tools and systems[1]–[3]. In this paper I focus on e-recruiting as one ofthe most important electronic-HRM (e-HRM) activities[3]. Existing e-recruiting Recommender Systems (RS)mainly focus on the problem of matching job demandsand candidates’ abilities. For the person-organizationalenvironment fit (P-OE fit)1 highly relevant social in-teractions between the potential employee and futurecolleagues at work (e.g. [4]–[18]) are not sufficientlytaken into account. To predict such interactions variousinputs are required (e.g. the organizational culture, theworkgroup’s social communication styles and means,a candidate’s personality and role behavior). A verypromising development in e-HRM is the incorporationof Online Social Networks (OSN). During the lastyears electronic recruiting in OSN (OSN-Recruiting)– envisioned a quarter of a century ago by Macdonald[19] – has emerged as a field of scientific research

1. The P-OE fit concerns all aspects of the fit between a jobcandidate and a vacancy within the recruiting organization, cf. 2.

[20]–[22]. But even these approaches only address thematching problem of job demands and a candidate’sabilities (person-job fit), which has become knownas the expert finding problem. However, rating theperson-job fit is just one part of the whole P-OEfit, e.g. [12,14]. Other crucial evaluation parts suchas the matching of the applicants’ personality to theorganizations’ culture or the candidates’ role behaviorto the workgroup fabric are largely ignored by bothInformation Systems (IS) research as well as practicale-recruiting solutions.

However, the underlying problems of these currentlywidely ignored social matching phenomena receivemuch more attention in sociology, psychology and theirrelated disciplines. Thus, incorporating sociologicalstudies such as Social Capital Theory [23]–[27] andSocial Network Analysis (SNA) [28]–[33], psychoso-cial studies on group dynamics [4]–[9], psychologicalinvestigations of personality traits and Internet/OSNuse [34]–[46] as well as cultural influences on In-ternet/OSN use [47]–[50], combining human resourcemanagement and occupational and vocational studies[10,12]–[15,51]–[57] into IS research is promising.

In this conceptual paper I sketch an interdisciplinaryframework for OSN-Recruiting based on an extensive,though by no means complete, review of the availableliterature. Implementing this framework will facilitaterecommendations to be made to recruiters. Further-more such a framework allows deeper interdisciplinaryinvestigations of the P-OE fit problem, which is long-standing and well-known in HRM.

The rest of the paper is organized as follows: Next Istudy candidate evaluation and selection from a HRMperspective in order to analyze the conditions for thebest P-OE fit. Then I show how relevant data can beextracted from OSN, before I take a preparatory stepfor designing OSN-Recruiting RS by mathematicallyformalizing the P-OE fit. I conclude with a discussionon the potential, limitations and the further applicationsof my framework.

Proceedings of the 47th Hawaii International Conference on System Sciences - 2014

This work is a preprint version. Copyright is held by IEEE.

2. Candidate evaluation and selection

A review of current empirical work on e-HRM canbe found in [1], and the most recent theory-based e-recruiting review has been provided by Maurer andCook [3]. In the latter study, the authors emphasizedthe key role of the P-OE fit in order to recruit highquality job applicants. Meta-analysis investigationsshowed significant positive correlations between the P-OE fit and job performance, job satisfaction, organiza-tional commitment and employee turnover [12]–[17].The P-OE fit can be broken-down into sub-fits, with themost appropriate consisting of three all but disjointedsub-fits, which together almost cover the whole notionof the P-OE fit [14]. These sub-fits are the Person-Organization (P-O) fit (between candidate personalityand organizational culture) [12], the Person-Group (P-G) fit (matching of individual and group roles andinteractions) [9] and the Person-Job (P-J) fit (betweena candidate’s skills, knowledge, and abilities and jobdemands) [11]. An overview of investigations intothese and further P-OE sub-fits can be found in [14].Thus we can formulate the following:

P-O fit ` P-G fit ` P-J fit ù P-OE fit . (1)

The relative importance of these sub-fits depends onthe concrete vacancy. Kristof-Brown [51] empiricallyshowed that recruiters actually distinguish betweenthese sub-fits. Sekiguchia and Huber [58] recentlyinvestigated the weighting of the P-O fit and theP-J fit when hiring decision-makers to evaluate jobcandidates. For instance the longer a recruiter expectsa candidate to stay within the organization the moreimportant the P-O fit becomes. The shorter the envi-sioned stay the more important the P-J fit becomes,since skill and knowledge acquisition on the job be-come inefficient for relatively short stays within theorganization. For a formalization see Eqn. (2).

2.1. Person-organization fit

The P-O fit describes the macro-perspective compat-ibility between the employee’s personalities and theorganization’s culture they work in [14, pp. 285]. Tomeasure the P-O fit several tests have been developed,e.g. [12]. The importance of such tests for hiringorganizations has been noted in [54, p. 21], and furtherresearch has been surveyed in [13]–[15,51,55]. Earlystudies of the P-O fit focused on the personality-climate congruence, while current investigations em-phasize the value congruence [12,13,56]. According tothese and other studies [10,55] employees are mostsuccessful in organizations with a culture which is

compatible with their personalities. Hence measuringthe P-O fit requires (a) an analysis of the candidate’spersonality and (b) an investigation of the organiza-tional culture:

Individual personality (part a) indicates a stablepattern of behavior [55]. Numerous approaches formodeling and measuring personality (traits) exist. TheFive Factor Model (FFM) refined by Costa and Mc-Crae [59] has been most widely and successfully used[60]. Within the FFM five personality traits (opennessto experience, conscientiousness, extraversion, agree-ableness, and neuroticism) describe an individual’spersonality and several ways to measure these traitshave been developed. The inventories NEO-FFI-3 andNEO-PI-R [59] are the most widely adopted.

To assess the organizational culture (part b) severalinstruments have been developed and a current litera-ture review of instruments for exploring and measuringorganizational culture is provided by Jung et al. [61] inwhich the authors identify 79 established instruments,of which 48 assess psychometric information (e.g. Hof-stede’s Culture Measures [62] and the OrganizationalCulture Profile (OCP) by O’Reilly et al. [63]).

2.2. Person-group fit

Interpersonal interactions within a workgroup (e.g.subordinates, coworkers and supervisors) are in certaincircumstances more important to team performancethan technical job skills [7,8]. The P-G fit concernssocial interaction aspects of the matching of a jobcandidate to the workgroup (meso-perspective). Thismatching has two parts, a supplementary one (similarqualities of group members) and a complementary one(distinctive qualities or characteristics) [9]. Measuringthe P-G fit requires (a) an analysis of the candidate’ssocial interaction characteristics and (b) an investiga-tion of the group’s social fabric.

Individual interaction characteristics are partly de-termined by personality traits [6], and these havebeen used to forecast team performance directly, cf.[6,7]. Mount et al. [6] showed by conducting a meta-analysis that agreeableness, emotional stability, andconscientiousness are positively correlated with teamperformance. Kristof-Brown et al. [53] revealed thatteams are more attracted to an individual with anextraversion score that is complementary to the groups’extraversion score (i.e. high individual-low team or lowindividual-high team levels).

Werbel and Johnson proposed a process to analyzethe workgroup [9, pp. 234]: (1) Investigate group in-teraction processes (e.g. time orientation, styles andmeans of communication: e-mail, phone or meetings)

Proceedings of the 47th Hawaii International Conference on System Sciences - 2014

This work is a preprint version. Copyright is held by IEEE.

and furthermore identify critical group norms (e.g.individualism or collectivism [62]). (2) The next stepis to investigate the performance of the team already inplace, as this allows the defining of the work role(s) tobe filled in order to improve team performance (com-plementary aspects). Critical task roles within a teamsuch as initiator-contributor, coordinator, elaborator,information seeker or information giver, and criticalmaintenance roles such as compromiser, encourageror follower were delineated by Benne and Sheats[4]; boundary spanning roles such as scout, ambas-sador, sentry and guard were identified by Anconaand Caldwell [5]. (3) The last step is to calculatethe P-G fit using both the supplementary (1) and thecomplementary (2) P-G fit.

2.3. Person-job fit

The P-J fit generally describes the micro-perspectivematching between a candidate’s characteristics (such asabilities or preferences) and job demands/supplies ofthe recruiting organization. Edwards [11] emphasizedtwo basic conceptualizations of the P-J fit. The so-called demands-abilities fit compromises the matchingof a candidate’s Skills, Knowledge and Abilities (SKA)on the one side and job requirements on the other side.The often labeled needs-supplies or supplies-values fitrefers to the question of whether the employee’s needs,desires or preferences are met by her job. The last typeof fit is mainly discussed within various theories ofadjustment, well-being, and satisfaction [14, pp. 284].Since these two conceptualizations can be seen as twosides of the same coin, I will only focus here on thedemands-ability fit.

To measure the P-J fit several tests have beendeveloped (e.g. P-J fit scale, Global Self-Report Mea-sure [52]). Following [12, p. 8] the P-J fit should beevaluated relative to the job tasks and not relative tothe employing organization. Hence, besides the mostwidely used inventories – which are questionnaire-based – direct examinations of a candidate’s CV, testi-monials and letters of reference together with the jobdescription – provided by the recruiting organization –can also be used to calculate the P-J fit.

In summary, I investigated in the above section theconditions for the best fit between a job candidate andthe recruiting organization and identified the triad ofP-O fit, P-G fit and P-J fit as crucial. These fits cor-respond to the well-known macro-, meso- and micro-perspective of SNA (Fig. 1).

Organization (macro)

Group (meso)

Person(micro)

Established linkPotential new link

Figure 1: Macro-, meso- and micro-perspective on the P-OE fit in OSN-Recruiting

3. Information extraction from online so-cial networks

In order to infer the job candidate’s personality traitsand SKA, the social fabric of the workgroup and theorganizational culture, the relevant data about OSNusers, their social environments and their interactionshave to be extracted. Recent work (e.g. [64]) showsthat the information stored in large-scale OSN (e.g.LinkedIn) is more truthful compared to simple manu-ally generated/maniplulated CVs.

3.1. Aspects of social network analysis

To address the problem of assessing the integrationof a person p in a social network (SN), connectivity andcentrality measures cf. [31,65,66] have been studiedintensively. One of the most basic measures to assessthe importance of a node is the degree centrality mea-sure of Shaw [28]. The degree centrality of a node n isdefined as the number of neighboring nodes. The close-ness centrality measure [29,30] focuses on the distancebetween nodes. Nodes contained in many short pathsbetween nodes can efficiently propagate informationor considerably impede the flow of information. Afurther classical notion of centrality is related to anode’s ability to influence the flow of information. Byway of illustration I mention the betweenness centralitymeasure [28,31], which is calculated as the number ofshortest paths a nodes is part of divided by the overallnumber of shortest paths in the network. Milgram hadalready demonstrated in 1967 that between almost anytwo nodes in an SN there is a short path linking

Proceedings of the 47th Hawaii International Conference on System Sciences - 2014

This work is a preprint version. Copyright is held by IEEE.

them [23]. Furthermore he showed that short pathscan be found without too much effort. These resultshave been found to also hold for OSN [67]. Althoughcomputing the distance between users in large OSNis an NP-hard problem, users can find short pathsconnecting them to other users [68,69]. Granovetter[24] showed that weak ties can be highly beneficialas they provide connections to “distant clusters”. Inparticular, weak ties are valuable for learning aboutvacancies [24, p. 1371]. Burt [26] did not focus onthe strength relationships but on the absence of linksin an OSN. He coined the term structural hole todescribe missing connections between groups in an SN.This notion proved to be highly influential and enabledmany important developments [27].

SNA has been applied to investigate individual per-formance as well as group performance, e.g. [70,71].The results show correlations between certain central-ity measures and job and group performance. Baldwinet al. [70] reported that communication centrality waslinked to student’s grades (i.e. performance), Sparroweet al. [71] found that group performance was negativelyrelated to hindrance network density.

3.2. Online social networks

OSN2 such as Facebook, LinkedIn or Xing havebecome a part of every day life.3 In turn OSN have al-ready become an important tool for recruiters [22,73].The scientific investigations of OSN [72]–[74] dateback about one and a half decade [75] and the focusof the research is diverse, including investigations intouser groups, use of OSN, data privacy and protection,social structures in OSN, graph theoretic properties ofOSN, comparisons between OSN and real-world SN.Data generated by OSN users has drawn considerablebusiness interest, of which [76] provides an overview.The value of social information stored in OSN, whichis key in my approach, has already been noted andexploited by the RS community [77].

To transfer insights from SNA gained by studyingnotions such as structural hole, degree and centralitymeasures to OSN, only the underlying graph structureof the OSN has to be known. To model weak andstrong links in OSN the notion of an activity networkhas been introduced. To distinguish between a good

2. The literature distinguishes in particular between OSN andSocial Network Sites (SNS). The former takes a sociological per-spective on electronic networks while the latter emphasizes the userinteraction with a website, cf. [72]. For my purposes both approachesare relevant, however I will here only use the term OSN.

3. LinkedIn reports having more than 200.000.000 users as ofApril 2013, Xing reports 13.000.000 users in December 2012 andFacebook round about 1.000.000.000 as of April 2013.

friend and a mere acquaintance the frequency of inter-actions can be measured. Although the activity networkof Facebook, made up of the regularly interactingusers, is a priori of a different shape and form thanthe whole network, Viswanath et al. [78] reported thatthe activity network graph-theoretically resembles thewhole network. In OSN weak ties also play a crucialrole for the transfer of knowledge [79,80], in particularfor job searches (cf. [24]).

3.3. Personality extraction from online socialnetworks

Inferring personality traits from Internet/OSN useis an emerging area of research. I will here highlightsome more relevant findings for our purposes by re-stricting our attention to the FFM. A good startingpoint is the work of Ryan et al. [45] containing awell researched state of the art overview on howpersonality traits influence behavior on Facebook andother OSN. Before going into details I remark thatthe hypothesis that OSN profiles are created to escapereality or to compensate for perceived personality flawshas been found to be in general false for Facebookusers [42,43]. These findings support the idea of per-sonality extractions from OSN. In fact, recruitmentand selection professionals regularly use OSN suchas LinkedIn and Facebook and they state that theyare able to draw conclusions on personality traitsfrom an applicant’s profile and network [22]. Goslinget al. [46] showed that Facebook-based personalityimpressions have some consensus and accuracy for allFFM dimensions, with particularly strong consensusfor extraversion, but with the exception of accuracyfor neuroticism.

Ross et al. [38] studied user behavior on Face-book via self-reports and found a weak connectionbetween the personalities of the users and their behav-ior. Furthermore they reported that extraversion wasnot significantly related to the number of Facebookfriends (in agreement with findings in [81]) nor wasthe time spent online thus related. But they found thatindividuals in the high Extraversion group reportedmembership in significantly more Facebook groups.Amichai-Hamburger and Vinitzky [39] replaced theself-reports of Ross et al. [38] by more objective crite-ria and found a strong connection between personalityand Facebook user behavior. Tong et al. [37] found aquartic relationship between the number of Facebookfriends of a user and perceptions of the individual’sextraversion. Kramer and Winter [36] investigated therelationship between self-reported (offline) personalitytraits and (online) self-presentation in the German OSN

Proceedings of the 47th Hawaii International Conference on System Sciences - 2014

This work is a preprint version. Copyright is held by IEEE.

StudiVZ. Their content analysis of the respondents’profiles showed that self-efficacy with regard to im-pression management (an aspect of extraversion) isstrongly related to the number of virtual friends, thelevel of profile detail, and the style of the profile photo.Extraverted users tended to present themselves in a lessrestrained manner choosing less conservative picturesof themselves. Recent studies (e.g. [40]–[42]) confirmthe strong link between extraversion and OSN usage.

Ross et al. [38] found a correlation between neu-roticism and the usage of Facebook apps. Individualswith a high neuroticism score were more likely todeclare the wall to be their favorite app whereas peoplewith a low score were more likely to prefer photos.Individuals with a high score in agreeableness werefound to be more likely to be selected as a friend, cf.[44]. Individuals that have high scores in openness toexperience and in neuroticism were found to be morelikely to be bloggers [35]. Karl et al. [48] analyzed therelationships between personality differences and fauxpas posts in Facebook. Their results showed that thosewho score high on conscientiousness, agreeablenessand emotional stability post significantly fewer fauxpas on their profile.

Interestingly the line of research assessing charactertraits of an individual given the profile and informationabout friends has been inverted. Denissen et al. [34]found that the user profiles of an individual’s friendscan be used to infer character traits of the individual.Results from [44] suggest that a friendship link be-tween users is more likely if they have similar scoresin agreeableness, extraversion and openness.

3.4. Organizational culture extraction

Organizational culture can either be measured manu-ally by conducting a questionnaire-based investigationof the organizations’ members or it can be inferredfrom their OSN use. The most common instrument formanual evaluation is the Organizational Culture Profile(OCP) developed by O’Reilly, Chatman and Caldwell[63]. The OCP is a Q-sort instrument containing 54value statements. The idea of an automated cultureextraction from OSN is supported by the fact thatthe organizational culture is reflected by the members’personalities and values [10,12,13,55,56]. The automa-tion of an OSN-based culture evaluation is promisingsince strong indications for culture dependent OSNuse have already been found (e.g. [47]–[50]). Yanget al. [50] showed that culture traits can be inferredfrom people’s social question and answer behavior inOSN. Pfeil’s et al. [47] findings suggest that culturaldifferences within the physical world also exist in the

virtual world. They show correlations between Hofst-ede’s [62] culture dimensions and use of Wikipedia(e.g. a correlation between the masculinity index andadding / clarifying information behavior). Callahan andHerring [49] also observed a systematic cultural bias inWikipedia use. Karl et al. [48] analyzed faux pas post-ing in Facebook profiles of US and German students.US students were more inclined than German studentsto post faux pas to their Facebook profile. The authorsdiscussed possible cultural reasons (specifically thatUS citizens tend to be more individualistic and lowerin uncertainty avoidance than Germans). In summarythere is a growing body of evidence to suggest that itis possible to automatically infer organizational culturetraits from OSN.

3.5. Communication style extraction from on-line social networks

Investigating communication styles from ComputerMediated Communications (CMC) is an emerging areaof research. In a first step it can be checked whetheran OSN profile is linked with further communicationtools such as Skype, ICQ, or Twitter. For instance,if an OSN profile is linked to a twitter account, itis possible to learn twitter use characteristics (e.g.messages sent, following users, other users followed,retweeting frequency and so on) [82]. Communicationintensity can be derived from blogging and tweetingintensity or wall use in Facebook [38]) and timeorientation can be inferred from online times.

Business interests in this area have also be-gun to emerge. The importance of an individual’scommunication style for recommending items toher has been noted and subsequently patented (US2011/0054985A1) by Cisco. Social Behavior Analysis(SBA) is the basis for two similar patterns granted toYahoo! that use SBA for generating recommendations(US 2009/0164400A1 and US 2009/0164897A1).

3.6. Role extraction from online social net-works

Since personality mainly predicts potential rolesof group members (e.g. [83]) we can infer theseroles from personality traits data extracted from OSN.As an example consider the “initiator-contributor”-role identified by Benne and Sheats [4] as a crit-ical task role in every group. This role is charac-terized by entrepreneurial attributes (such as sugges-tions of new ideas and solutions to the group, (re-)definition of group problems and goals, organizingthe group for the task ahead, handling difficulties the

Proceedings of the 47th Hawaii International Conference on System Sciences - 2014

This work is a preprint version. Copyright is held by IEEE.

group encounters, etc. [4, p. 43]). Zhao et al. [83] meta-analytically summarized the relationship of personalitytraits and these entrepreneurial attributes. A broadrange of personality scales were categorized into aset of constructs using the FFM of personality. Theirresults showed that four of the five FFM dimensionswere associated with entrepreneurial intentions andentrepreneurial performance. The ability to fill the“information seeker” role [4] could be inferred forinstance from Facebook use [84] or from followingnumerous Twitter users, while similarly the “harmo-nizer” role can be inferred from the use of the groupfeature [84].

3.7. Extraction of skills, knowledge and abili-ties from online social networks

Measuring SKA has been a hot scientific topic[11,12,51,52,58,85]. HR-XML is a current commonstandard for expressing SKA in machine-readable for-mat. Today HR-XML is extended and maintained bya non-profit consortium and has also made its wayinto practical applications in real world labor markets.For instance, the German employment agency (BA)has adopted this format and developed a specificationcalled HR-BA-XML. It is my firm view that it willonly be a matter of time before (business) OSN adopt agenerally accepted fully machine-readable standard foruploading CVs, testimonials and letters of reference.

The problem of finding a person possessing certainSKA (i.e. an expert) to solve a concrete problem or taskhas scientifically been studied and is known under thelabel expert finding or expert location. SN have provedto be a fertile ground to search for experts. Searchstrategies for expert finding in SN were evaluated in[21] and the state of the art has been reviewed in [86].

3.8. Legal aspects of data extraction fromonline social networks

In practice we cannot assume that all informationstored in OSN is freely available. In general acces-sibility of data is limited by user settings, by dataprotection and privacy laws and the OSN terms andconditions. Projecting current trends suggests that theselaws are only set to tighten. For the foreseeable futurehowever I believe that (future) laws do not renderOSN-Recruiting unfeasible. I foresee two possible ap-proaches to maintaining recruiting RS searching OSNdata even under tightened rules and regulations. Oneis that privacy-preserving data mining techniques [87]can be applied, see [76] for a current state of the artoverview of data anonymisation techniques). Secondly,

micropayments can be made to OSN users and/or theOSN provider for the right to extract data.

It is possible that current trends will reverse andusers of OSN will explicitly want their profiles tobe public to further business opportunities and careeradvancement. It is plausible to conjecture that users ofbusiness OSN (e.g. LinkedIn and Xing) will prefer tofollow the latter trend.

If and how personalities may be measured andsubsequently legally used in a recruiting process is acomplex legal question. Different laws (in particularthe ever changing labor and anti-discrimination laws)and legal systems have given different answers whichvary by country and also over time, cf. [88].

4. Recommender systems for online socialnetwork recruiting

The design of (OSN-)Recruiting RS is an emerg-ing area, cf. [20,21,77,85]. But the existing RS onlyaddress the P-J fit with the exception of Malinowskiet al. [85] who took both the P-J and the P-G fit intoaccount. However they completely ignored the P-O fitand focused only “on a company-internal team staffingscenario” [85, p. 431].

Here I consider the design of content-based andcollaborative RS for OSN-Recruiting. Content basedRS estimate the utility a user obtains from an item idepending on other items the user has previously ratedwhich are similar to item i. Collaborative approachesdiffer from the above by recommending items to auser u based on ratings or recommendations by userswhich are similar to user u. It is feasible to designcontent-based recruiting RS searching OSN based ona similarity notion between candidates and an assess-ment of selected candidates. Such collaborative RS canbe designed using recruiting strategies from organiza-tions similar to the recruiting organization, and suchstrategies can be obtained from (industrial) espionageor less clandestinely by monitoring job ads and portals,or, more in line with my topic, by monitoring OSN.Given that content based and collaborative approachesare possible hybrid RS are conceivable.

Before an RS can recommend candidates whichhave not been assessed by the user the problem ofhow a given candidate is evaluated has to be ad-dressed. Formally such a utility function is given bythe P-OE fit, which can be represented by a functionFP�OE . FP�OE takes as inputs a vacancy V withinan organization and a candidate C and returns a realnumber. This function may be decomposed into threesub-functions: fP�O, fP�G and fP�J (representingthe P-O, the P-G and the P-J fit) and an aggregation

Proceedings of the 47th Hawaii International Conference on System Sciences - 2014

This work is a preprint version. Copyright is held by IEEE.

function A. A combines these sub-functions by anaggregation of the relative importance placed on thesub-fits, cf. 2. We can thus formulate the following:

FP�OEpC, V q � ApfP�OpC, V q, fP�GpC, V q, fP�J pC, V qq(2)

Determining the aggregation function A constitutesa Multi Criteria Decision Making (MCDM) problem.One main challenge in MCDM is the above com-mensurateness problem, i.e. if, and if so how, a lowscore in one criterion can be compensated by a highscore in another criterion. The Choquet and the Sugenointegrals [89] are methods that are widely adoptedby the MCDM community to address this problem.The RS community is well aware of multi-criterialphenomena and has developed RS based on multipleevaluation criteria [90].

Do note that the P-G fit only depends on certainaspects of a vacancy. Indeed only the social fabricof the workgroup SFWGV is relevant while forinstance required skills are completely irrelevantfor determining fP�G. Similarly not all propertiesof a candidate C matter for calculating fP�G, asonly the candidate’s social interaction patterns SIPC

are required. We can hence formulate fP�GpC, V q �

fP�GpfcompP�G pSIPC , SFWGV q, fsupp

P�GpSIPC , SFWGV qq

where f compP�G , fsupp

P�G are the complementaryrespectively supplementary P-G fit. Likewise theP-O fit only depends on the organization’s culturalvalues VV and the candidate’s personality PC whereasthe P-J fit only depends on job requirements JRV anda candidate’s skills, abilities and knowledge SKAC .We hence reformulate equation (2) as

FP�OEpC, V q � ApfP�OpPC , VV q, (3)

fP�GpfcompP�G pSIPC , SFWGV q, fsupp

P�GpSIPC , SFWGV qq,

fP�J pSKAC , JRV qq .

The challenges posed to the RS community com-promise more than the development of RS estimatingFP�OEpC, V q based on equation (3). Due to the highcomplexity of modern work environments recruitingdecisions are made under uncertainty [57]. It appearsreasonable to use a structure that relies on the sameformal model of uncertainty for all three fits, whichmight be phrased in quantitative (probabilistic), quali-tative or linguistic (fuzzy) terms. Figure 2 graphicallysummarizes the proposed framework.

5. Conclusions

I have developed an interdisciplinary framework forOSN-Recruiting. Adopting a business HRM perspec-

Social context

Integrated Perspective

Micro levelMeso levelMacro level

CandidatePersonality*Poten-

tial Roles*

Organizational Culture*

Group Roles*

P-O fit P-G fit

Group Com-muni-cation styles*

Com-muni-cation style*

Skills, Knowledge& Abilities*

Job Description

P-J fit

Hiring Organization

P-OE fit

* Data inferable from OSN

Figure 2: An interdisciplinary P-OE fitbased framework for OSN-Recruiting

tive I identified the P-OE fit as crucial in the assess-ment of the suitability of job candidates to a givenvacancy. As shown, the P-OE fit can be broken downinto the P-O fit, the P-G fit and the P-J fit, cf. Eqn. (1).In contrast to previous approaches I hence includedin my framework the macro-perspective (P-O fit) andthe meso-perspective (P-G fit). For the calculation ofall these sub-fits I identified the relevant inputs (e.g.personality, organizational culture) and I showed thatthese inputs can potentially be inferred from OSN withthe sole exception of the job description. Finally I tooka preparatory step for designing OSN-Recruiting RSby mathematically formalizing the P-OE fit (Eqn. (3)).This equation can then be used to forecast the P-OEfit between an OSN user and a vacancy.

My main contributions are (a) illustrating a way toimprove existing recruiting RS and (b) providing aplatform for interdisciplinary research thereby enablingfurther discoveries in various areas of research.

5.1. Limitations

The business potential of my framework for OSN-Recruiting appears enormous, however there are sev-eral limitations: (i) There are sociopolitical, legaland psychological barriers/reservations concerning dis-closing, mining and processing of personal informa-tion. Privacy concerns can be addressed via privacy-preserving data mining technologies or by marketmechanisms for data extraction from OSN in return forrewards, cf. 3.8. (ii) Inferring social characteristics oforganizations, groups and individuals based on data ex-tracted from OSN has not yet matured into a broad andconsistent area of research. However recent researchconverges, for instance see 3.3 for extraversion. (iii)My framework is limited by the models I used, such

Proceedings of the 47th Hawaii International Conference on System Sciences - 2014

This work is a preprint version. Copyright is held by IEEE.

as the P-OE fit model, the FFM, workgroup modelsetc. However I used the most accepted and advancedmodels to represent the state of the art, cf. 2. Furtherdevelopments of the underlying models can be usedto enhance my framework. (iv) Finally at present myframework is curtailed due to a missing evaluation (e.g.by prototypes, field studies, tests), cf. 5.2.

5.2. Future research

For a successful development and implementation ofan OSN-Recruiting RS based on my framework sev-eral interesting open interdisciplinary problems needto be addressed. In particular I identify: (i) study-ing organizational culture extraction from OSN, (ii)mining OSN for the social fabric of a workgroup(group role detection, communication style extraction),(iii) inferring personality traits based on OSN data,(iv) improving models predicting on-the-job behaviorbased on data extracted from OSN, cf. [48, p. 182], (v)designing and implementing a fully machine-readablestandard for CVs, letters of references and testimonials,(vi) developing mechanisms for trading OSN data,(vii) implementing and evaluating of real-world OSN-Recruiting RS, and (viii) studying of users of such RS.

Completing this research program does not onlysupports recruiters but may also yield a better un-derstanding of the P-OE fit problem. This deepenedunderstanding enables further contributions to sociol-ogy (group dynamics, SNA), psychology (personalityresearch, OSN use), computer science (data miningand design of OSN/RS) and economics (organizationalbehavior research).

Finally my framework may also be applied to otherHRM problems such as organizational development,personnel development and retention management.

Acknowledgments

I would like to thank Juergen Landes for his supporton an earlier version of the paper. This research ispartly funded by the German Federal Ministry of Edu-cation and Research (BMBF, contract 03FH055PX2).

References

[1] S. Strohmeier, “Research in e-HRM: Review and im-plications,” HRMR, vol. 17, no. 1, pp. 19–37, 2007.

[2] I. Lee, “Modeling the benefit of e-recruiting processintegration,” DSS, vol. 51, no. 1, pp. 230–239, 2011.

[3] S. D. Maurer and D. P. Cook, “Using company websites to e-recruit qualified applicants: A job marketingbased review of theory-based research,” Comput HumBehav, vol. 27, no. 1, pp. 106–117, 2011.

[4] K. D. Benne and P. Sheats, “Functional Roles ofGroup Members,” J Soc Issues, vol. 4, no. 2, pp.41–49, May 1948.

[5] D. G. Ancona and D. F. Caldwell, “Beyond Task andMaintenance: Defining External Functions in Groups,”Group & Organization Studies, vol. 13, no. 4, pp. 468–494, December 1988.

[6] M. K. Mount, M. R. Barrick, and G. L. Stewart,“Five-Factor Model of personality and Performancein Jobs Involving Interpersonal Interactions,” HumanPerformance, vol. 11, no. 2-3, pp. 145–165, 1998.

[7] S. T. Bell, “Deep-level composition variables as pre-dictors of team performance: A meta-analysis,” J. Appl.Psychol., vol. 92, no. 3, pp. 595–615, May 2007.

[8] J. R. Mesmer-Magnus and L. A. DeChurch, “In-formation Sharing and Team Performance: A Meta-Analysis,” J. Appl. Psychol., vol. 94, no. 2, pp. 535–546, 2009.

[9] J. D. Werbel and D. J. Johnson, “The Use of Person-Group Fit for Employment Selection: A Missing Linkin Person-Environment Fit,” Hum. Resour. Manage.,vol. 40, no. 3, pp. 227–240, 2001.

[10] V. R. Tom, “The Role of Personality and OrganizationalImages in the Recruiting Process,” Organ. Behav. Hum.Perform., vol. 6, no. 5, pp. 573–592, 1971.

[11] J. R. Edwards, Person-job fit: A conceptual integra-tion, literature review, and methodological critique.International review of industrial and organizationalpsychology. Oxford, England: John Wiley & Sons,1991, vol. 6, pp. 283–357.

[12] A. L. Kristof, “Person-organization fit: An integrativereview of its conceptualizations, measurement, and im-plications,” Pers Psych, vol. 49, no. 1, pp. 1–49, 1996.

[13] M. L. Verquer, T. A. Beehr, and S. H. Wagner, “Ameta-analysis of relations between person-organizationfit and work attitudes,” J. Vocat. Behav., vol. 63, no. 3,pp. 473–489, 2003.

[14] A. L. Kristof-Brown, R. D. Zimmerman, and E. C.Johnson, “Consequences of individuals fit at work:A meta-analysis of person-job, person-organization,person-group, and person-supervisor fit,” Pers Psych,vol. 58, no. 2, pp. 281–342, 2005.

[15] B. J. Hoffman and D. J. Woehr, “A quantitative reviewof the relationship between person-organization fit andbehavioral outcomes,” J. Vocat. Behav., vol. 68, no. 3,pp. 389–399, 2006.

[16] A. Leung and S. Chaturvedi, “Linking the fits, fittingthe links: Connecting different types of PO fit toattitudinal outcomes,” J. Vocat. Behav., vol. 79, no. 2,pp. 391–402, 2011.

[17] A. Ramesh and M. J. Gelfand, “Will They Stay or WillThey Go? The Role of Job Embeddedness in PredictingTurnover in Individualistic and Collectivistic Cultures,”J. Appl. Psychol., vol. 95, no. 5, pp. 807–823, 2010.

[18] J. E. Sheridan, “Organizational Culture and EmployeeRetention,” Acad Manage J, vol. 35, no. 5, pp.1036–1056, 1992.

[19] S. Macdonald, “Headhunting in high technology,” Tech-novation, vol. 4, no. 3, pp. 233–245, 1986.

[20] T. Keim, “Extending the Applicability of RecommenderSystems: A Multilayer Framework for MatchingHuman Resources,” in HICSS 2007 Proceedings,2007, p. 169.

Proceedings of the 47th Hawaii International Conference on System Sciences - 2014

This work is a preprint version. Copyright is held by IEEE.

[21] J. Zhang and M. S. Ackerman, “Searching for Expertisein Social Networks: A Simulation of Potential Strate-gies,” in GROUP ’05 Proc. ACM, 2005, pp. 71–80.

[22] R. Caers and V. Castelyns, “LinkedIn and Facebook inBelgium: The Influences and Biases of Social NetworkSites in Recruitment and Selection Procedures,” SocSci Comput Rev, vol. 29, no. 4, pp. 437–448, 2011.

[23] S. Milgram, “The small world problem,” PsychologyToday, vol. 2, no. 1, pp. 60–67, 1967.

[24] M. S. Granovetter, “The Strength of Weak Ties,” Am JSociol, vol. 78, no. 6, pp. 1360–1380, 1973.

[25] J. S. Coleman, “Social Capital in the Creation ofHuman Capital,” Am J Sociol, vol. 94, no. Supplement:Organizations and Institutions: Sociological andEconomic Approaches to the Analysis of SocialStructure, pp. 95–120, 1988.

[26] R. S. Burt, Structural holes: The social structure ofcompetition. Harvard University Press, 1992.

[27] G. Walker, B. Kogut, and W. Shan, “Social Capital,Structural Holes and the Formation of an IndustryNetwork,” Organ Sci, vol. 8, no. 2, pp. 109–125, 1997.

[28] M. E. Shaw, “Group Structure and the Behavior ofIndividuals in Small Groups,” J Psychol, vol. 38, no. 1,pp. 139–149, 1954.

[29] M. A. Beauchamp, “An Improved Index of [Centrality,”Behav. Sci., vol. 10, no. 2, pp. 161–163, April 1965.

[30] G. Sabidussi, “The centrality index of a graph,”Psychometrika, vol. 31, no. 4, pp. 581–603, 1966.

[31] L. C. Freeman, “Centrality in Social Networks:Conceptual Clarification,” Social Networks, vol. 1,no. 3, pp. 215–239, 1978-1979.

[32] S. Wasserman and K. Faust, Social Network Analysis:Methods and Applications. Cambridge UniversityPress, 1994.

[33] J. P. Scott, Social Network Analysis: A Handbook.Sage, 2007.

[34] J. J. A. Denissen, R. Geenen, M. Selfhout, and M. A. G.van Aken, “Single-item Big Five Ratings in a SocialNetwork Design,” Eur J Personality, vol. 22, no. 1, pp.37–54, 2008.

[35] R. E. Guadagno, B. M. Okdie, and C. A. Eno, “Whoblogs? Personality predictors of blogging,” ComputHum Behav, vol. 24, no. 5, pp. 1993–2004, 2008.

[36] N. Kramer and S. Winter, “Impression Management2.0: The Relationship of Self-Esteem, Extraversion,Self-Efficacy, and Self-Presentation Within Social Net-working,” J Media Psych, vol. 20, no. 3, pp. 106–116,2008.

[37] S. T. Tong, B. Van Der Heide, L. Langwell,and J. B. Walther, “Too Much of a GoodThing? The Relationship Between Number of Friendsand Interpersonal Impressions on Facebook,” JCMC,vol. 13, no. 3, pp. 531–549, 2008.

[38] C. Ross, E. S. Orr, M. Sisic, J. M. Arseneault,M. G. Simmering, and R. R. Orr, “Personality andmotivations associated with Facebook use,” ComputHum Behav, vol. 25, no. 2, pp. 578–586, 2009.

[39] Y. Amichai-Hamburger and G. Vinitzky, “Social net-work use and personality,” Comput Hum Behav, vol. 26,no. 6, pp. 1289–1295, 2010.

[40] T. Correa, A. W. Hinsley, and H. G. de Zuniga,“Who interacts on the web?: The intersection of

users’ personality and social media use,” Comput HumBehav, vol. 26, no. 2, pp. 247–253, 2010.

[41] E. Y. Ong, R. P. Ang, J. C. Ho, J. C. Lim, D. H. Goh,C. S. Lee, and A. Y. Chua, “Narcissism, extraversionand adolescents’ self-presentation on Facebook,” PersIndivid Dif, vol. 50, no. 2, pp. 180–185, 2011.

[42] S. D. Gosling, A. A. Augustine, S. Vazire, N. Holtz-man, and S. Gaddis, “Manifestations of Personalityin Online Social Networks: Self-Reported Facebook-Related Behaviors and Observable Profile Information,”Cyberpsychol Behav Soc Netw, vol. 14, no. 9, pp. 483–488, 2011.

[43] M. D. Back, J. M. Stopfer, S. Vazire, S. Gaddis,S. C. Schmukle, B. Egloff, and S. D. Gosling, “Face-book Profiles Reflect Actual Personality, Not Self-Idealization,” Psychol Sci, vol. 21, no. 3, pp. 372–374,2010.

[44] M. Selfhout, W. Burk, S. Branje, J. Denissen, M. vanAken, and W. Meeus, “Emerging Late AdolescentFriendship Networks and Big Five Personality Traits:A Social Network Approach,” J. Pers., vol. 78, no. 2,pp. 509–538, 2010.

[45] T. Ryan and S. Xenos, “Who uses Facebook? aninvestigation into the relationship between the BigFive, shyness, narcissism, loneliness, and Facebookusage,” Comput Hum Behav, vol. 27, no. 5, pp.1658–1664, 2011.

[46] S. D. Gosling, S. Gaddis, and S. Vazire, “PersonalityImpressions Based on Facebook Profiles,” in ICWSM’07 Proc. AAAI, 2007, pp. 1–4.

[47] U. Pfeil, P. Zaphiris, and C. S. Ang, “CulturalDifferences in Collaborative Authoring of Wikipedia,”JCMC, vol. 12, no. 1, pp. 88–113, 2006.

[48] K. Karl, J. Peluchette, and C. Schlaegel, “Who’sPosting Facebook Faux Pas? A Cross-CulturalExamination of Personality Differences,” Int J SelectAssess, vol. 18, no. 2, pp. 174–186, 2010.

[49] E. S. Callahan and S. C. Herring, “Cultural biasin Wikipedia content on famous persons,” JASIST,vol. 62, no. 10, pp. 1899–1915, 2011.

[50] J. Yang, M. R. Morris, J. Teevan, L. A. Adamic, andM. S. Ackerman, “Culture Matters: A Survey Study ofSocial Q&A Behavior,” in ICWSM ’11 Proc. AAAI,2011, pp. 409–416.

[51] A. L. Kristof-Brown, “Perceived applicant fit:Distinguishing between recruiters’ perceptions ofperson-job and person-organization fit,” Pers Psych,vol. 53, no. 3, pp. 643–671, 2000.

[52] M. Brkich, D. Jeffs, and S. A. Carless, “A GlobalSelf-Report Measure of Person-Job Fit,” Eur J PsycholAssess, vol. 18, no. 1, pp. 43–51, 2002.

[53] A. Kristof-Brown, M. R. Barrick, and C. Kay Stevens,“When Opposites Attract: A Multi-Sample Demon-stration of Complementary Person-Team Fit onExtraversion,” J. Pers., vol. 73, no. 4, pp. 935–958,2005.

[54] D. M. Cable and C. K. Parsons, “Socialization tacticsand person-organization fit,” Pers Psych, vol. 54, no. 1,pp. 1–23, 2001.

[55] C. Anderson, S. E. Spataro, and F. J. Flynn, “Person-ality and Organizational Culture as Determinants ofInfluence,” J. Appl. Psychol., vol. 93, no. 3, pp. 702–710, 2008.

Proceedings of the 47th Hawaii International Conference on System Sciences - 2014

This work is a preprint version. Copyright is held by IEEE.

[56] J. R. Edwards and D. M. Cable, “The Value of ValueCongruence,” J. Appl. Psychol., vol. 94, no. 3, pp.654–677, 2009.

[57] A. M. Spence, “Job Market Signaling,” Q J Econ,vol. 87, no. 3, pp. 355–374, 1973.

[58] T. Sekiguchi and V. L. Huber, “The use of person-organization fit and person-job fit information inmaking selection decisions,” Organ. Behav. Hum.Decis. Process., vol. 116, no. 2, pp. 203–216, 2011.

[59] P. T. Costa and R. R. McCrae, Revised NEO person-ality inventory (NEO-PI-R) and the NEO Five-Factorinventory (NEO-FFI): Professional manual. Odessa,FL, USA: PAR, 1992.

[60] M. G. Rothstein and R. D. Goffin, “The use ofpersonality measures in personnel selection: Whatdoes current research support?” HRMR, vol. 16, no. 2,pp. 155–180, 2006.

[61] T. Jung, T. Scott, H. T. O. Davies, P. Bower,D. Whalley, R. McNally, and R. Mannion, “Instrumentsfor Exploring Organizational Culture: A Review ofthe Literature,” Public Adm. Rev., vol. 69, no. 6, pp.1087–1096, 2009.

[62] G. H. Hofstede, Culture’s Consequences: InternationalDifferences in Work-Related Values. Beverly Hills,CA: Sage, 1984.

[63] C. A. O’Reilly III, J. Chatman, and D. F.Caldwell, “People and Organizational Culture: AProfile Comparison Approach to Assessing Person-Organization Fit,” Acad Manage J, vol. 34, no. 3, pp.487–516, 1991.

[64] J. Guillory and J. T. Hancock, “The Effect of Linkedinon Deception in Resumes,” Cyberpsychol Behav SocNetw, vol. 15, no. 3, pp. 135–140, 2012.

[65] L. C. Freeman, “A Set of Measures of Centrality Basedon Betweenness,” Sociometry, vol. 40, no. 1, pp. 35–41,1977.

[66] S. P. Borgatti and M. G. Everett, “A Graph-theoreticperspective on centrality,” Social Networks, vol. 28,no. 4, pp. 466–484, 2006.

[67] P. S. Dodds, R. Muhamad, and D. J. Watts, “An Exper-imental Study of Search in Global Social Networks,”Science, vol. 301, no. 5634, pp. 827–829, 2003.

[68] L. Adamic and E. Adar, “How to search a socialnetwork,” Social Networks, vol. 27, no. 3, pp. 187–203,2005.

[69] D. J. Watts, P. S. Dodds, and M. E. J. Newman,“Identity and Search in Social Networks,” Science,vol. 296, no. 5571, pp. 1302–1305, 2002.

[70] T. T. Baldwin, M. D. Bedell, and J. L. Johnson,“The Social Fabric of a Team-Based M.B.A.Program: Network Effects on Student Satisfactionand Performance,” Acad. Manage. J., vol. 40, no. 6,pp. 1369–1397, 1997.

[71] R. T. Sparrowe, R. C. Liden, S. J. Wayne, and M. L.Kraimer, “Social Networks and the Performance ofIndividuals and Groups,” Acad. Manage. J., vol. 44,no. 2, pp. 316–325, 2001.

[72] d. m. boyd and N. B. Ellison, “Social Network Sites:Definition, History, and Scholarship,” JCMC, vol. 13,no. 1, pp. 210–230, 2008.

[73] W. P. Smith and D. L. Kidder, “Youve been tagged!(Then again, maybe not): Employers and Facebook,”Bus. Horiz., vol. 53, no. 5, pp. 491–499, 2010.

[74] D. Rosen, G. A. Barnett, and J.-H. Kim, “Socialnetworks and online environments: when science andpractice co-evolve,” Social Network Analysis andMining, vol. 1, no. 1, pp. 27–42, 2011.

[75] L. Garton, C. Haythornthwaite, and B. Wellman,“Studying Online Social Networks,” JCMC, vol. 3,no. 1, 1997.

[76] F. Bonchi, C. Castillo, A. Gionis, and A. Jaimes,“Social Network Analysis and Mining for BusinessApplications,” ACM TIST, vol. 2, no. 3, 2011, ArticleNo. 22.

[77] D. Jannach, M. Zanker, A. Felfernig, and G. Friedrich,Recommender Systems: An Introduction. CambridgeUniversity Press, 2010.

[78] B. Viswanath, A. Mislove, M. Cha, and P. K.Gummadi, “On the evolution of user interaction inFacebook,” in WOSN ’09 Proc., 2009, pp. 37–42.

[79] D. Z. Levin and R. Cross, “The Strength of WeakTies You Can Trust: The Mediating Role of Trust inEffective Knowledge Transfer,” Management Science,vol. 50, no. 11, pp. 1477–1490, 2004.

[80] J. Donath, “Signals in Social Supernets,” JCMC,vol. 13, no. 1, pp. 231–251, 2007.

[81] T. Amiel and S. L. Sargent, “Individual differences inInternet usage motives,” Comput Hum Behav, vol. 20,no. 6, pp. 711–726, 2004.

[82] C. Wang and B. Huberman, “Long trend dynamics insocial media,” EPJ Data Science, vol. 1, no. 1, p. 2,2012.

[83] H. Zhao, S. E. Seibert, and G. Lumpkin, “The Re-lationship of Personality to Entrepreneurial Intentionsand Performance: A Meta-Analytic Review,” Journal ofManagement, vol. 36, no. 2, pp. 381–404, March 2010.

[84] D. Y. Wohn, C. Lampe, J. Vitak, and N. Ellison,“Coordinating the ordinary: social information uses ofFacebook by adults,” in iConference ’11 Proc., 2011,pp. 340–347.

[85] J. Malinowski, T. Weitzel, and T. Keim, “Decisionsupport for team staffing: An automated relationalrecommendation approach,” DSS, vol. 45, no. 3, pp.429–447, 2008.

[86] T. Lappas, K. Liu, and E. Terzi, “A Survey ofAlgorithms and Systems for Expert Location in SocialNetworks,” in Social Network Data Analytics, C. C.Aggarwal, Ed. Springer, 2011, pp. 215–241.

[87] R. Agrawal and R. Srikant, “Privacy-Preserving DataMining,” in SIGMOD ’00 Proc. ACM, 2000, pp.439–450.

[88] R. L. Hogler, C. Henle, and C. Bemus, “Internetrecruiting and employment discrimination: a legalperspective,” HRMR, vol. 8, no. 2, pp. 149–164, 1998.

[89] M. Grabisch and C. Labreuche, “A decade ofapplication of the Choquet and Sugeno integralsin multi-criteria decision aid,” Annals of OperationsResearch, vol. 175, no. 1, pp. 247–286, 2010.

[90] G. Adomavicius, N. Manouselis, and Y. Kwon, “Multi-criteria recommender systems,” in RecommenderSystems Handbook. Springer, 2011, pp. 769–803.

Proceedings of the 47th Hawaii International Conference on System Sciences - 2014

This work is a preprint version. Copyright is held by IEEE.