Title:
Discovering Social Circles in Directed Graphs

Keywords:
Social circles

Abstract:
<p>We examine the problem of identifying social circles, , or sets of cohesive and mutually aware nodes surrounding ... the value of including a particular node in an emerging social circle, , and describe a greedy social circle discovery algorithm. We demonstrate the effectiveness of this approach on ...

References:
C. Kadushin. 1966. The friends and supporters of psychotherapy: On social circles in urban life. American Sociological Review 31, 6 (1966), 786--802. http://www.jstor.org/stable/2091658.
C. Kadushin. 1968. Power, influence and social circles: A new methodology for studying opinion makers. American Sociological Review 33, 5 (1968), 685--699. http://www.jstor.org/stable/2092880.
J. McAuley and J. Leskovec. 2012. Learning to discover social circles in ego networks. In Advances in Neural Information Processing Systems 25, P. Bartlett, F. C. N. Pereira, C. J. C. Burges, L. Bottou, and K. Q. Weinberger (Eds.). 548--556. http://books.nips.cc/papers/files/nips25/NIPS2012_0272.pdf.
H. Qin, T. Liu, and Y. Ma. 2012. Mining user’s real social circle in microblog. In Proceedings of the IEEE ACM International Conference on Advances in Social Networks Analysis and Mining. 348--352.

Full Text:
TKDD0804-2121Discovering Social Circles in Directed GraphsSCOTT H. BURTON and CHRISTOPHE G. GIRAUD-CARRIER, Brigham ... G. GIRAUD-CARRIER, Brigham Young UniversityWe examine the problem of identifying social circles, , or sets of cohesive and mutually aware nodes surround-ing ... the value of including a particular node in an emerging social circle, ,and describe a greedy social circle discovery algorithm. We demonstrate the effectiveness of this approachon artificial ... I.2.m.[Artificial Intelligence]: MiscellaneousGeneral Terms: Algorithms, PerformanceAdditional Key Words and Phrases: Social circles, , directed graphs, local community searchACM Reference Format:Scott H. Burton ... Reference Format:Scott H. Burton and Christophe G. Giraud-Carrier. 2014. Discovering social circles in directed graphs. ACMTrans. Knowl. Discov. Data 8, 4, Article ...
... such as cowork-ers, teammates, or fellow hobbyists. For example, the social circle of an individual thatbegins with two of his or her ... her sisters is likely to center around family relationships,while the social circle of that same individual that begins with two of his ... that sense, suchlocal communities resemble what sociologists refer to as social circles [Kadushin 1966,1968], and we refer to them as such in ... in the following.Here, we focus our attention on the local social circle discovery problem. Analogous tothe difference between community detection and local ... community detection and local community detection, the localvariant of the social circle discovery problem operates under the constraint that theentire graph is ... approach clearly inadequate.In this article, we propose an effective local social circle discovery algorithm fordirected graphs. Ideally, seed nodes are selected such ... to thesocial circle are iteratively added, or those in the social circle are periodically removed,by maximizing a particular heuristic function until a ... Vol. 8, No. 4, Article 21, Publication date: August 2014.Discovering Social Circles in Directed Graphs 21:3affiliation [Yang and Leskovec 2012a]. A recent, ...
... affect the algorithm or metric. By contrast, we propose alocal social circle discovery algorithm for directed graphs.Recently, McAuley and Leskovec [2012] have ... [2012] have presented a generative, unsupervisedapproach to discover an individual?s social circles among their friends, which combineslink and profile information, while Qin ... first, within the computer science research community, touse the term social circle. . Their definition is similar to ours since they ?expect ... a small set of its altersonly) and build a single social circle around them. One significant distinction is thatthe resulting social circle may contain individuals who are not directly connected to ego(i.e., ... arealready connected [Faloutsos et al. 2004; Tong and Faloutsos 2006].3. SOCIAL CIRCLE DISCOVERYThe local social circle discovery problem for directed graphs consists of identifying aset of ...
... Vol. 8, No. 4, Article 21, Publication date: August 2014.Discovering Social Circles in Directed Graphs 21:5has twice as many edges as Graph ... which is composed of only unidirectional links.Even more relevant to social circle discovery, since all of the edges in Graph 4 pointdownward, ... there seems to be a significant semanticdifference in terms of social circle membership between a node with high in-degree(e.g., a news site ... return to the specific issues raised by the localdiscovery of social circles in that context.3.1. The Lab Advisor ProblemSince a social circle is defined as a cohesive group of nodes around an ... the decision to include a new node in a given social circle shouldbe independent from the existence of other collateral social circles to which that nodemay also belong [Friggeri et al. 2011].As ... al. 2011].As an example, consider the task of discovering the social circle around a few studentswho work in the same research lab. ... work in the same research lab. One would expect that social circle to encompass allstudents in the lab, as well as the ... A?s students (query set), the lab should here bethe discovered social circle. . Note that many community mining algorithms that viewa community ... fewer total number of connections.In the case of naturally overlapping social circles, , the fundamental assumption of?more in than out? does not ... connections? [Ahn et al. 2010]. A node?s membershipto a specific social circle should depend solely on the strength of its ties to ...
node to add to an existing social circle, , the denominatoris the same for all candidates, since in ... the number of links of the candidate node to the social circle. . Hence, tomaximize Densityd, one only needs to maximize its ... numerator, E. Now, E counts thenumber of edges in the social circle so that for all candidate nodes, E starts at the ... and e(x, y) = 0 otherwise. Let SC be a social circle and n a node that may beadded to SC. Then, ... dd(n, SC)would allow A to be added to the lab social circle. . Similarly, as expected, if the setof query nodes were ... Vol. 8, No. 4, Article 21, Publication date: August 2014.Discovering Social Circles in Directed Graphs 21:7Fig. 3. The problem of adding nodes ... circle.community, rather than a few of her students, the resulting social circle would includeA and her colleagues, but none of her students. ... needed, however, in responseto other important issues.3.2. The Fringe ProblemA social circle is defined as a cohesive group of nodes that surround ... that, ideally, its nodescapture the characteristic that defines the desired social circle. . As a result, one wouldexpect the query set to ... the area labeled 2 beginto be added to the growing social circle, , they will have a tendency to cause the highlyconnected ... the area labeled X to be part of the final social circle, , rather than the nodesin the dense area marked 3. ... time, based only on information from nodes in the growing social circle. . If newermembers of the social circle are treated the same as older ones, then they exert ... influence on new ones and can thus easily cause the social circle to divert fromthe initial query set, as illustrated earlier. Hence, ... the idea ofdiscounted importance through length of membership to the social circle, , as follows.ACM Transactions on Knowledge Discovery from Data, Vol. ... Giraud-CarrierTo maintain the relative importance of early members of the social circle, , especiallythe query set, edges between nodes are discounted according ... according to the time when thenodes were included in the social circle. . Let s(n) denote the step in the social circle- -building process at which node nwas added to the social circle. . We modify Equation (2)to obtain the step-discounted value ?d(n, ...
process less and less important to the growing social circle. . Smaller values of? reduce the impact of when nodes ... which weighs thevalue of connections to a node in the social circle inversely to the step in which it wasadded, thereby giving ... and the distinction between in-degree andout-degree and its impact on social circle membership.3.3. The Famous Person ProblemWe have already addressed the issues ... setcentrality. There remains as part of the definition of a social circle the fact that itshould be composed of nodes that are ... each other person? [Shaw 1976]. Whilewe do not require a social circle to be a k-clique, it is reasonable to expect that ... k-clique, it is reasonable to expect that eachmember of the social circle influences and is influenced by at least some other membersof ... a potential node to have links from everymember of the social circle if there are no links back, and vice versa.As an ... There likely exist in such a graph dense groups, or social circles, ,of respected research scientists who have cited each other?s work ... (i.e., linking to them) does not makeher part of their social circle in any meaningful way. In the second, and somewhatreciprocal case, ... Vol. 8, No. 4, Article 21, Publication date: August 2014.Discovering Social Circles in Directed Graphs 21:9Fig. 4. Two types of nodes that ... abstractly in Figure 4,where the shaded nodes mark the current social circle, , Node A represents the newresearcher in the first instance, ... directed edges. Node A links to several nodes in the social circle, , but thereare no links from members of the circle ... it. Conversely, Node B has links fromseveral members of the social circle but does not link back to any of them. Neitherone ... them. Neitherone of these nodes should be part of the social circle. . To help exclude such nodes andenforce some level of ... some level of mutual influence as per the definition of social circles, , we makeone final change to the node selection function, ... ?(n, SC) over all candidate nodes, we ensure that the social circle isdense, centered around the initial query set, and its members ...
by nodes that link to the social circle, , but of which the algorithm iscurrently unaware (due to ... a 0, because such nodes have no links from the social circle, , and thus they can beconsistently excluded. Only those nodes ... beconsistently excluded. Only those nodes that have links from the social circle can have ascore greater than 0 (the min term would ... can be safely reduced to only those that are known.3.4. Social Circle Discovery AlgorithmSozio and Gionis [2010] have proven that a greedy ... greedy algorithm, equippedwith the function ?, is guaranteed to solvethe social circle discovery problem. However, as onemay expect, that algorithm, and thesubsequent ... the problem, where only thosenodes that members of the growing social circle link to are available. While we cannotguarantee global optimality in ... and the frequency of removal, and produces as output a social circle of at mostthe specified size. Two global variables, AddStep and ... new node is to be considered for addition tothe growing social circle (line 8) and decremented every time a node is removed ... 8) and decremented every time a node is removed fromthe social circle ... (line 22). When a node w is added to the social circle, , its s(w) valueis initialized to the current value of ... (line 14), and when a node c is removedfrom the social circle, , the value s(x) of each node x that was ... to avoid skipping add-step values (lines 19?21).Lines 1?5 initialize the social circle SC to the set of query nodes. The algorithm thengoes ... Vol. 8, No. 4, Article 21, Publication date: August 2014.Discovering Social Circles in Directed Graphs 21:11ALGORITHM 1: Social Circle Discovery AlgorithmInput: Set Q of initial query nodes, maximum size ... Q of initial query nodes, maximum size max of the social circle, , and frequency ofnode removal fOutput: A social circle SC of size at most max1: AddStep ? 12: for ... NumIter ? NumIter+ 125: end while26: Return SCinclusion in the social circle
... or the set of all nodes that anymember of the social circle links to (line 9). If N is empty, then there ... N is empty, then there is no way to growthe social circle any further and the algorithm simply breaks out of the ... the value of ? is selected andadded to the current social circle (lines 13?15). In the event of a tie, the tie ... removal, as described earlier. If so, the node in thecurrent social circle with the smallest ? value is selected and removed from ... them. Once the algorithm comes outof the loop, the final social circle SC, of size at most max, is returned (line 26).3.4.1. ... (line 26).3.4.1. Stopping Criterion. One important aspect of the local social circle discovery prob-lem is the decision of when to stop growing ... be applied. In thislatter case, the mechanisms for building a social circle and the decision to stop it areACM Transactions on Knowledge ... a universal stopping criterion for local community search. Like others,our social circle discovery algorithm does not have a built-in stopping criterion. Ourcurrent ... work.3.4.2. Query Set Selection. Another important aspect of the local social circle discoveryproblem is the selection of the query set (parameter Q ... the query set is independent of the mechanicsof discovering a social circle. . Indeed, given any query set, Q, Algorithm 1 will ... what determines the desiredsocial circle. Thus, the quality of a social circle depends heavily on the nodes in Qexhibiting and sharing the ... in Qexhibiting and sharing the characteristics that define the desired social circle, , so that,in practice, the choice of Q is critical. ... inour experiments, our focus here is directed primarily at identifying social circles oncea query set has been defined.3.4.3. Computational Complexity. There are ...
... Vol. 8, No. 4, Article 21, Publication date: August 2014.Discovering Social Circles in Directed Graphs 21:13so that the complexity of computing ?(n, ... of nodes N = 1,000.For each configuration setting, a separate social circle is discovered around each ofthe 1,000 nodes as the initial ...
... However, in this case, we are interestedin finding a separate social circle around every node in the graph. Thus, we compareonly the ... of the algorithms are stopped when the size of the social circle matches the size ofthe correct community defined in the benchmark ... the correct community has 25members, the algorithms run until the social circle contains at most 25 nodes). In thecase of the Greedy ... of a node, only the outgoingneighbors are returned. Once a social circle is discovered, it is compared against thecorrect community by evaluating ... the F-measure. A separate F-measure value is deter-mined for the social circle around each start node and then averaged across the 1,000circles. ... Vol. 8, No. 4, Article 21, Publication date: August 2014.Discovering Social Circles in Directed Graphs 21:15Fig. 6. Comparison on overlapping communities.process, they ... CommunitiesNext, we evaluate the ability of each algorithm to discover social circles in graphswhere nodes may belong to multiple overlapping communities. For ... overlapping nodes to be either two orfour communities.As earlier, separate social circles are discovered around each of the 1,000 nodes inthe graph. ... in each desired community. The results wereagain averaged across all social circles discovered in the graph with the mean andstandard deviations of ...
... Query Node SelectionAs discussed, an important element of the local social circle discovery problem is theselection of query nodes. While in many ... set is determined beforehandand is the reason for discovering the social circle, , in other cases, it may be that an initialmember ... be added to the query set to discover the desired social circle. . For example, considerthe case of identifying a social circle of business contacts around an individual. Thequestion arises, which coworkers ... Vol. 8, No. 4, Article 21, Publication date: August 2014.Discovering Social Circles in Directed Graphs 21:17Fig. 7. Number of correct nodes found ... usethe two nodes as a query set to discover a social circle. . We treat the network as if onlyoutgoing connections are ... impact the effectiveness of discovering other nodesin the desired local social circle. . Using the best method of query node selection earlier(highest ... method of query node selection earlier(highest in-group ratio), we discover social circles around query sets that range in sizefrom two to 10 ...
... using more than two nodes helps to better identify the social circle ofinterest, with the largest increase in value coming from adding ... to identify, so by starting with them already in the social circle, , the taskis to find other, potentially more ?difficult,? members.5.3. ... select a second node for the query set anddiscover a social circle around them. For the selection of the second node, we ... trials completed. Here, each trialrepresents the discovery of a separate community/social circle from different query sets.For example, 100 trials means that 100 ... represents the average number of correct nodes added to 500separate social circles. . Note also that, as stated earlier, we focus strictly ... Vol. 8, No. 4, Article 21, Publication date: August 2014.Discovering Social Circles in Directed Graphs 21:19Fig. 9. Number of correct nodes added ... 2014.21:20 S. H. Burton and C. G. Giraud-CarrierTable I. NBA Social Circle MembersStep Twitter Account Name NBA Team1. KingJames LeBron James Miami1. ...
... Twitter User Social CirclesWe first apply our approach to building social circles of users on the social networkplatform Twitter. For demonstration purposes, ... 2013.We first turn to professional basketball players and discover a social circle aroundthree prominent players: LeBron James, NBA player for the Miami ... 62).By choosing players from different cities, we avoid discovering a social circle focusedon a certain market such as radio or TV personalities ... Using these three accounts as the queryset, we apply our Social Circle Discovery algorithm to build a social circle of max = 75members, with ? = 1 and f ... in Table I. Of the 75 members of the discovered social circle, , 62 were NBAplayers or groups, four were affiliated with ... it is actually possible to be interested in, anddiscover, other social circles or overlapping communities. Indeed, in addition to beinga professional basketball ... Vol. 8, No. 4, Article 21, Publication date: August 2014.Discovering Social Circles in Directed Graphs 21:21Table II. Popular Culture Social Circle MembersStep Twitter Account Name/Stage Name Status1. KingJames LeBron James Athlete1. ... Musician17. DwyaneWade Dwyane Wade Athlete18. DJCLUE DJ Clue? Musicianthat another social circle may be obtained if we include popular figures rather thanprofessional ... set with him. To verify this hypothesis andfurther validate our Social Circle Discovery algorithm, we rerun the algorithm with aquery set comprising ... request rate restrictionsenforced by Twitter. The members of the resulting social circle ... are listed in Table II.All of the members of this social circle are entertainers of some kind, and each oftheir Twitter accounts ... four are professional athletes. This group clearly representsa rather different social circle to which Lebron James also belongs. Incidentally, hisfriendship with the ...
... party. Choosing fiveDemocrats and five Republicans, we build two separate social circles of 100 members(i.e., max = 100), with ? = 1 ... ? 22K; following: ?8K)Each of the members of the discovered social circles were involved in politics, eventhough not all of them were ... no representatives of the opposite political party were discoveredin the social circles, , suggesting little direct overlap among members.ACM Transactions on Knowledge ... Vol. 8, No. 4, Article 21, Publication date: August 2014.Discovering Social Circles in Directed Graphs 21:23Table IV. The First 25 Members of ...
connections between commonblogs can define social circles within the blogosphere, and because the entire set ofblogs cannot ... discovery method is required to discover thesesets. To discover a social circle of blogs, we downloaded the latest 50 blog entries foreach ... meta-data, it was considered a blog. A casestudy of blog social circles complements that of Twitter social circles nicely, becausewhereas on Twitter many graphs are densely connected and ... for manyusers to follow those that follow them, a blog social circle defined by links to other blogsis much more sparse.A prominent ... children and homemaking and linkto one another. To discover a social circle around a set of mommy blogs, we selectedthe query set ... 2014.21:24 S. H. Burton and C. G. Giraud-CarrierFig. 11. The social circle of mommy blogs. The larger nodes are the initial query ... as larger nodes) remains highly connected andprominent in the resulting social circle, , as opposed to being left on the fringe while ... AND FUTURE WORKIn this article, we have defined the local social circle discovery problem in directedgraphs and proposed a novel algorithm to ... in directedgraphs and proposed a novel algorithm to discover such social circles around an ini-tial query set, based on a degree-inspired quality ... that quantifies the value ofadding a node to the growing social circle. . Our approach does not focus on boundariesand can therefore ... focus on boundariesand can therefore include appropriate nodes in a social circle regardless of their mem-bership in other circles. In addition, it ... leaving the initial query nodes onthe fringe of the final social circle. . Further, our approach explicitly accounts for edgedirection and avoids ... or unknown nodes that do not have mutualinteraction with the social circle. . We show that our Social Circle Discovery algorithmperforms well on artificial benchmark problems, on large networks ... in real-world networks. Our method is ableto efficiently discover meaningful social circles even when the degree of the includednodes is extremely high.ACM ... Vol. 8, No. 4, Article 21, Publication date: August 2014.Discovering Social Circles in Directed Graphs 21:25There are several interesting extensions to our ... to study themost effective query sets to uniquely identify the social circle
... 103018.C. Kadushin. 1966. The friends and supporters of psychotherapy: On social circles in urban life. AmericanSociological Review 31, 6 (1966), 786?802. http://www.jstor.org/stable/2091658.C. ... 31, 6 (1966), 786?802. http://www.jstor.org/stable/2091658.C. Kadushin. 1968. Power, influence and social circles: : A new methodology for studying opinion makers.American Sociological Review ...
... (2008), 387?400.J. McAuley and J. Leskovec. 2012. Learning to discover social circles in ego networks. In Advances in NeuralInformation Processing Systems 25, ... Vol. 8, No. 4, Article 21, Publication date: August 2014.Discovering Social Circles in Directed Graphs 21:27V.Nicosia, G.Mangioni, V. Carchiolo, andM.Malgeri. 2009. Extending ... Qin, T. Liu, and Y. Ma. 2012. Mining user?s real social circle in microblog. In Proceedings of the IEEE/ACMInternational Conference on Advances ...

Title:
Automatic Social Circle Detection Using Multi-View Clustering

Keywords:
social circles

Abstract:
... assign friends into circles.</p> <p>In this paper, we introduce a social circle discovery approach using multi-view clustering. First, we present our observations ... First, we present our observations on the key features of social circles: : friendship links, content similarity and social interactions. We propose ... techniques. Results show that multi-view clustering is more accurate for social circle detection; and our proposed approach gains significantly higher similarity ratio ...

References:
J. McAuley and J. Leskovec. Discovering social circles in ego networks. TKDD'13, 2012.
A. Squicciarini, S. Karumanchi, D. Lin, and N. DeSisto. Identifying hidden social circles for advanced privacy configuration. Computers & Security, 2013.

Full Text:
Automatic Social Circle Detection Using Multi-ViewClusteringYuhao Yang, Chao Lan, Xiaoli Li, Bo Luo, ... manually assign friends into circles.In this paper, we introduce a social circle discovery ap-proach using multi-view clustering. First, we present ourobservations on ... clustering. First, we present ourobservations on the key features of social circles: : friendshiplinks, content similarity and social interactions. We proposea one-side ... clustering techniques. Results show that multi-viewclustering is more accurate for social circle detection; andour proposed approach gains significantly higher similarityratio than the ... message owners have full control of the informa-tion boundary. Meanwhile, social circles are also expectedto promote information sharing, since they give users ...
... or too few links (sometimes referred-to as?hubs? and ?outliers?). Meanwhile, social circle identifica-tion approaches from the user-centered research communityoften require explicit attributes, ... this paper, we present a multi-view clustering approachto automatically discover social circles in users? ego net-works. Besides the topology-based clusters adopted in ... interaction fea-tures will improve clustering performance, and eventuallygenerate more meaningful social circles. . We notice thatsome views are very sparse (e.g. the ... first to in-tegrate structural, content and interaction features to iden-tify social circles in online social networks; (2) we introducea novel selective co-trained ... the nodes that connect to the user, i.e., allhis/her friends. Social circles of a user?s ego network arehidden structures of closely connected ...
... also indicates a close relation-ship. In summary, we identify three social circles from thisexample: {(A,B,C); (D,E); (G,F )}. As we can see, ... share many friends in common.Observation 2. Users from the same social circle tend toshare interests on similar contents and opinions.Observation 3. Users ... of information from users? ego networks to automat-ically identify non-overlapping social circles. . We define sixviews that belong to three categories to ...
... network. The next step is to integratethese views to identify social circles. .3. MULTI-VIEW CLUSTERING3.1 Notations and OperatorsWe use capital letter to ...
... not directly penalize imbal-ance cluster results, since in applications some social circles, ,such as family, are indeed smaller than other circles, suchas ...
... better than CSC.4.5 Manual EvaluationUltimately, the quality of the discovered social circles mustbe assessed by users. To include users in the loop, ... perceive the separations of differentcircles in the ego network.5. RELATEDWORKIdentifying social circles from a user?s online social net-works is important for the ...
... 47].Based on privacy concerns, [29] developed a model to dis-cover social circles by using both network structure and userprofile information; [40] proposed ... more easily.Algorithmically, we employ the multi-view clustering frame-work to detect social circles considering the multi-view na-ture of an ego network. This framework ...
... could be converted to two non-overlapping circles A;B A.Applications of social circles. . As suggested in [40, 41],social circles are used to protect information privacy, by de-livering messages to ... In socialization, messages are posted to the se-lected circles. Meanwhile, social circle enforcement becomesparticularly challenging when some social networking sitesallows breaches in ... the scope of this paper.On the other hand, the discovered social circles could beused to improve the efficiency of ad delivery, targeted ... we can extend our model to furtherstudy the development of social circles and evolution of egonetworks.7. CONCLUSIONWith the extreme popularity of online ... information and to facilitate secure socializa-tion. However, the problem of social circle discovery remainsopen and challenging. In this paper, we start with ...
... A. Squicciarini, S. Karumanchi, D. Lin, andN. DeSisto. Identifying hidden social circles foradvanced privacy configuration. Computers &Security, 2013.[41] A. Squicciarini, D. Lin, ...

Title:
Provably Fast Inference of Latent Features from Networks: with Applications to Learning Social Circles and Multilabel Classification

Keywords:
social circles

Abstract:
... efficiently high quality labelings. We use our method to learn social circles

References:
J. Leskovec and J. J. Mcauley. Learning to discover social circles in ego networks. In NIPS, pages 539--547, 2012.

Full Text:
... Fast Inference of Latent Features from Networkswith Applications to Learning Social Circles and Multilabel ClassificationCharalampos E. TsourakakisHarvard School of Engineering and Applied ... learns efficiently high quality labelings. We use ourmethod to learn social circles from Twitter ego-networks andperform multilabel classification.Categories and Subject DescriptorsG.2.2 [Graph ...
... For details, see text of Sections 1and 3.est include learning social circles [29, 43], distributed com-putation [7], detecting multifunctional proteins [12], min-ing ...
... This value for Wresults in a normalized version of Objective 1.Social circles. . Using our rule-of-thumb we apply ourMetropolis chain adapted for ...
... method is faster than BigClam for10 values of p.4.5 Learning social circles from ego-networksIn this Section we focus on a special type ...
et al. consider is the following:can we recover the social circles of a user in social mediabased on the structure of ... number of latent features is set equal to the numberof social circles. . BigClam is a successful method for findingoverlapping communities and ... BigClam is a successful method for findingoverlapping communities and detecting social circles [44].The results are shown in Table 3. The first two ... 0.2258 479 1.00 1.00 0.002 1.00 1.00 0.16Table 3: Learning social circles from ego-networksfrom Twitter results. For details, see text of Sec-tion ... cliques. Mcauleyand Leskovec use overlapping community detection meth-ods to discover social circles from ego-networks [29]. Asubsequent method developed by Yang and Leskovec ...
... method toperform successfully multilabel classification from similaritygraphs and to learn social circles from Twitter ego-networks.Open Problems. Our work opens numerous questions.What are ...
... Method4 Experimental results4.1 Experimental Setup4.2 Experimental findings4.3 Tiles4.4 Classification4.5 Learning social circles

Title:
Discovering social circles in ego networks

Keywords:
social circles

Abstract:
... networking sites allow users to manually categorize their friends into social circles (e.g., &ldquo;circles&rdquo; on Google&plus;, and &ldquo;lists&rdquo; on Facebook and Twitter). ... article, we study the novel task of automatically identifying users' social circles. . We pose this task as a multimembership node clustering ...

References:
J. McAuley and J. Leskovec. 2012. Learning to discover social circles in ego networks. In Neural Information Processing Systems.

Full Text:
TKDD0801-044Discovering Social Circles in Ego NetworksJULIAN MCAULEY and JURE LESKOVEC, Computer Science Department, ... networking sites allow users to manually categorize their friends into social circles( (e.g., ?circles? on Google+, and ?lists? on Facebook and Twitter). ... this article, we study the novel taskof automatically identifying users? social circles. . We pose this task as a multimembership node clusteringproblem ... ClusteringGeneral Terms: Algorithms, ExperimentationAdditional Key Words and Phrases: Community detection, social circles, , ego networks, machine learningACM Reference Format:Julian McAuley and Jure ... machine learningACM Reference Format:Julian McAuley and Jure Leskovec. 2014. Discovering social circles in ego networks. ACM Trans. Knowl.Discov. Data 8, 1, Article ... generated by them is to categorize their friends intosocial circles. Social circles are lists of people that can be used for content ...
... Vol. 8, No. 1, Article 4, Publication date: February 2014.Discovering Social Circles in Ego Networks 4:32005], and many circles are hierarchically nested ...
... SOCIAL CIRCLESIn the following discussion, we describe our model of social circles. . We desire a modelof circle formation with the following ...
February 2014.Discovering Social Circles in Ego Networks 4:5school; the corresponding dimension of ?k would ... profiles to suit our particular application.We describe a model of social circles that treats circle memberships as latent vari-ables. Nodes within a ... similarityparameters to best explain the observed network data.Our model of social circles is defined as follows. Given an ego network G and ...
... Vol. 8, No. 1, Article 4, Publication date: February 2014.Discovering Social Circles in Ego Networks 4:7where xi and xj are boolean variables, ...
... Vol. 8, No. 1, Article 4, Publication date: February 2014.Discovering Social Circles in Ego Networks 4:9In other words, an additional circle will ...
... Vol. 8, No. 1, Article 4, Publication date: February 2014.Discovering Social Circles in Ego Networks 4:11Fig. 2. Histogram of overlap between circles ...
... Vol. 8, No. 1, Article 4, Publication date: February 2014.Discovering Social Circles in Ego Networks 4:13obtain high-quality hand-labeled data, which we have ...
... representation scheme.We propose two hypotheses for how users organize their social circles: : either theymay form circles around users who share some ... Vol. 8, No. 1, Article 4, Publication date: February 2014.Discovering Social Circles in Ego Networks 4:15Fig. 4. Features obtained from Facebook. Each ...
... Vol. 8, No. 1, Article 4, Publication date: February 2014.Discovering Social Circles in Ego Networks 4:17This measure assigns equal importance to false ...
... Vol. 8, No. 1, Article 4, Publication date: February 2014.Discovering Social Circles in Ego Networks 4:19Table I. Methods for Circle and Community ...
... Vol. 8, No. 1, Article 4, Publication date: February 2014.Discovering Social Circles in Ego Networks 4:21Fig. 6. Top: Three detected circles on ...
... Vol. 8, No. 1, Article 4, Publication date: February 2014.Discovering Social Circles in Ego Networks 4:23Fig. 9. Number of seeds versus accuracy ...
... Vol. 8, No. 1, Article 4, Publication date: February 2014.Discovering Social Circles in Ego Networks 4:25Fig. 11. Accuracy of our MCMC algorithm ...
... labeled data that may beavailable in practice. Experiments reveal that social circles can be accurately detectedusing a combination of both network and ... Vol. 8, No. 1, Article 4, Publication date: February 2014.Discovering Social Circles in Ego Networks 4:27J. Chang and D. Blei. 2009. Relational ...
... University Press.J. McAuley and J. Leskovec. 2012. Learning to discover social circles in ego networks. In Neural InformationProcessing Systems.M. McPherson. 1983. An ...

Title:
An unsupervised learning approach to social circles detection in ego bluetooth proximity network

Keywords:
social circle learning

Abstract:
... a user is socially related into what we call as social circles. . In this note, we propose a novel unsupervised approach ... recording one's dynamic proximity relations with others to identify her social circles, , each of which is formed along a semantically coherent ...

References:
J. McAuley and J. Leskovec. Learning to discover social circles in ego networks. In Proc. NIPS 2012, pages 548--556, 2012.

Full Text:
An unsupervised learning approach to social circles detection in ego bluetooth proximity networkAn Unsupervised Learning Approach to ... whom a user is socially related into what wecall as social circles. . In this note, we propose a novel unsu-pervised approach ... senseddata recording one?s dynamic proximity relations with othersto identify her social circles, , each of which is formed alonga semantically coherent aspect. ... mobile data in sensingpersonal behaviors beyond personal data.Author KeywordsBluetooth Sensing; Social Circle Learning; Human BehaviorAnalysis; Collaborative FilteringACM Classification KeywordsI.5 Pattern Recognition; H.5.2 ... study a user?s social interactions is to identifyand characterize her social circles. . This concept is broadlyused in online social networks for ... promising type of data that can beleveraged to discover one?s social circles in physical worldis the time-stamped proximity data continuously sensed byBluetooth ... long periods of time.A user in reality often has multiple social circles, , each ofwhich consists of a subset of members in ... time and show its power in un-covering and separating meaningful social circles hidden in auser?s personal proximity network. Intuitively, a user?s inter-action ...
... dynamic proximity network as subjects. Ourtask then is to discover social circles from a user?s ego prox-imity network, including for each circle ... seemingly peculiar PTS and hence be wronglyclassified to an idiosyncratic social circle; ; multiple subjectswho are in the same circle may end ... and t in a proximity recordare potentially correlated via a social circle semantics as a la-tent factor, which in spirit is similar ... via latent interest semantics in recommenda-tion systems. Recall that a social circle is driven by a latentaspect and is characterized by its ... subject s and atime slot t ?match? in the latent social circle semantics, then itis highly likely that s is observed proximate ...
... potentially share similar circle membership. Afterlearning, the temporal characteristics of social circle k is en-coded in the kth row of matrix Q, ... allow multi-circle member-ship of a subject) is to ?force? different social circles to formalong different temporal dimensions, which makes sense asan ego?s ... dimensions, which makes sense asan ego?s interactions with her different social circles seldomoverlap in time (e.g, ?colleague? and ?family? interactionsSession: Social Computing ... regularization, each row in Qaccounts for a single semantically coherent social circle as-pect, which is what we desire. In addition, we add ... semantics, based on the intuition that the activetime for a social circle interaction is usually piecewise con-tinuous rather than sporadic. The final ... can be used for verification.Qualitatively, we first show what ego-centric social circles arelearned and explain by results how our model overcomes thementioned ...
... Q as random smallFigure 1. Temporal profiles of 4 typical social circles of ego 6each circle. Figure 1 clearly shows 4 typical ... assumed as prior knowledge. We next turnour focus to the social circle membership of the subjects topositive values, such that the gradient ... 8?12, 2013, Zurich, Switzerland723Social circle idSubject id1 2 310020921103222615104899307391?2?1012Figure 2. Social circle membership for selected subjectswhom user 6 is socially related. Such ...
... based on whether thesubject and the given time match in social circle semantics, aknowledge that is learned by exploiting all subjects? histori-cal ... method gives superior results,due to its awareness of the underlying social circle semanticsand its collaborative nature. But the key advantage of ourmethod ... WORKIn this note we proposed a learning framework to discoverego-centric social circles from BT-sensed proximity data,based on the intuition that it is ... from check-in logs. Another extension is to modelhow a user?s social circles evolve over time, which can offercrucial insights into users? social ... ACM, 2012.6. J. McAuley and J. Leskovec. Learning to discover social circles in egonetworks. In Proc. NIPS 2012, pages 548?556, 2012.7. J. ...

Title:
Social circle analysis on ego-network based on context frequent pattern mining

Keywords:
social circle

Abstract:
... difficult to promote. In this paper, an automatic identification of social circles model is proposed, which takes three factors into account: user's ... a user in ego-network can belong to more than one social circles or even no one. The experiments show that our model ... one. The experiments show that our model can effectively detect social circles

References:
J. McAuley and J. Leskovec. Discovering social circles in ego networks. arXiv preprint arXiv:1210.8182, 2012.

Full Text:
Social circle analysis on ego-network based on context frequent pattern miningSocial Circle ... di?cult to promote. In this paper, an automatic identi-?cation of social circles model is proposed, which takes threefactors into account: user's pro?le, ... a user in ego-network can belong to more than one social circles or even noone. The experiments show that our model can ... even noone. The experiments show that our model can e?ectivelydetect social circles on ego-network.Categories and Subject DescriptorsH.3.4 [Systems and Software]: Information networkGeneral ... and selectively transmit information.Similar groups and circles are de?ned as 'social circle' ' byMicrosoft, and unlike community, such social circle focuseson individual choice. That is to say, one individual canbe ... well. Figure1 shows the complex socialcircle phenomenon in ego-network.Although the social circle in social network has been triedto be analyzed, there is ... is no automatic and complete methodto explain the formation of social circle when the social circleis found.In this paper, the discovery and ... Department friends Single Center User Figure 1: Complex phenomenon of social circles
... ofexcavation so as to achieve the purpose of increasing e?-ciency.2.3 Social Circle AnalysisMicrosoft[13] ?rstly introduced the concept of social cir-cle, and modeled ... the concept of social cir-cle, and modeled the generation of social circle by setting aprobabilistic model, and individuals in social circle are au-tomatically divided by the method of machine learning andparameter ... considers only the impact ofindividual attribute to the formation of social circle, , but notcares about the inruence of interaction between individualsto ... about the inruence of interaction between individualsto internet construction and social circle formation.Mo Tong et.al.[14] proposed a perception method for micro-blog community ... is de?ned, and closed frequentitemsets are found to analysis the social circles in one ego-network.3. INTERACTION OF MICRO-BLOG3.1 Interactive RelationDivided the behavior ...
... However, only one interaction between users maybe random and uncertain.The social circle in the micro-blog and the reason of itsformation can be ... all micro-blog content of the author.Hypothesis 2: the formation of social circle among micro-blog users is rerected in the interaction of users. ... onehand, frequent interaction is the possible reason of forma-tion of social circle; ; on the other hand, social circle formsinteraction. Intuitively, there are generous set intersectionsbetween interactive context between ... interactive context between two individuals in thesame social circle.De?nition 5. social circle: : based on hypothesis 2, we canconsider that a social circle M = (u1; u2; :::; um) in user'sego-network meets Inter(u1) ... = A;A 6= ;.Users in the sameM are in one social circles ,and Inter(u10)Inter(u20) ::: Inter(un0) = A is their common contextcharacteristics.To ... ::: Inter(un0) = A is their common contextcharacteristics.To ?nd these social circles, , we proposed a framework tomine all social circles in one's ego-network , more detail canbe shown in Part ... one's ego-network , more detail canbe shown in Part 4.4. SOCIAL CIRCLE ANALYSIS4.1 FrameworkAfter determining the analysis user(the center user) inmicro-blog service, ... extract frequent itemsets withclearer semantic features, then we ?nd the social circles com-bining the frequent itemsets which is obtained in the last ... mining process with ego-network of the user. Theprocess framework of social circle analysis in detail is shownin ?gure2.141Ego network Users? profile Micro ... interactive context feature Closed frequent itemsets Mining frequent itemsets Reduce Social circle Analysis Figure 2: The process of social circle analysis4.2 Context Characteristics ExtractThere are two broad approaches to present ...
... forming the subset. So mining frequentpatterns can not only divide social circles on ego-network,but also can give direct explanation of formation mechanism.But ... in CT;32: end if33: end for34: end for35: return CT.4.4 Social Circle AnalysisFiltered through the minimum support, there will be alot of ... on de?nition def, we give social discoveryalgorithm as algorithm2.Algorithm 2 Social Circles Discovery Algorithm1: Input: Frequent ItemSets F, Friendship G.2: Output: Super ...
... micro-blogis 32102.Part experimental results are shown as Table1.There are two social circle samples discovered by our algo-rithm.For example, the id of social circle is 1, which has 28members. For privacy protection, we do ... we do not show the nick-name of the members. In social circle 1, all the membersare interested in data mining, so we ... social cir-cle is based on common jobs or interests. In social circle 2,all the members are interested in food and Changsha(whichis a ... Changsha(whichis a city name in China). We can ?nd that social circle 2is based on delicious food in local city. The result ... Table1, we ?nd node 4 and node19 are both in social circle 1 and social circle 2, so the socialcircles can be overlap by using our ... frequent itemsets obtainedusing di?erent process.010203040503 7 15 30Days Num. of social circles Figure 4: The number of social circles obtained us-ing the data of di?erent days.that some nodes are ... of di?erent days.that some nodes are not members of any social circles. . Thereason is that the users have not published enough ... of socialcircles found reached on the steady state.6. CONCLUSIONSIdentifying the social circles in one's ego-network and un-derstanding the formation of the social circles can improvemicro-blog value-added services,information management andthe understanding of the dissemination ...