finding cohesive subgroups and relevant members in the nokia friend view mobile social network
DESCRIPTION
Presentation from the Social Intelligence and Networking workshop at the IEEE Social Computing conference in Vancouver, Canada on Sunday, August 30, 2009TRANSCRIPT
Alvin Chin – Social Intelligence and Networking workshop, IEEE SocialCom’09
Finding Cohesive Subgroups and Relevant Members in the Nokia Friend View Mobile Social Network
Alvin Chin
Member of Research StaffMobile Social Networking GroupNokia Research Center, BeijingAugust 30, 2009
Alvin Chin – Social Intelligence and Networking workshop, IEEE SocialCom’09
Agenda1. Motivation2. Problem3. Solution4. Finding cohesive subgroups and important
members5. Case study: Nokia Friend View6. Results7. Conclusion and Future Work
Alvin Chin – Social Intelligence and Networking workshop, IEEE SocialCom’09
Many people want to be my “friend”
Who should I add to my social network as friends?
Alvin Chin – Social Intelligence and Networking workshop, IEEE SocialCom’09
How can we find who are the relevant members in
the social network?
Alvin Chin – Social Intelligence and Networking workshop, IEEE SocialCom’09
Solution• Create method for finding cohesive subgroups
and their relevant members based on SCAN method (Chin and Chignell, 2008)
• Apply it to Nokia Friend View network to see its viability and to describe its social behaviour
Alvin Chin – Social Intelligence and Networking workshop, IEEE SocialCom’09
Agenda1. Motivation2. Problem3. Solution4. Finding cohesive subgroups and important
members5. Case study: Nokia Friend View6. Results7. Conclusion and Future Work
Alvin Chin – Social Intelligence and Networking workshop, IEEE SocialCom’09
Cohesive subgroups• Group of members having more and frequent
interactions with each other than outside group
Clique K-plex
12
7
10
6 8
A B
C D
15A
E
B
F
Alvin Chin – Social Intelligence and Networking workshop, IEEE SocialCom’09
Finding cohesive subgroups• How
• Social network analysis, centrality, clustering • But no automated or systematic method• My work: modified SCAN method adapted from
Chin and Chignell (2008)
Chin, Alvin and Chignell, Mark(2008). Automatic detection of cohesive subgroups within social hypertext: A heuristic approach, New Review of Hypermedia and Multimedia,14:1, 121 — 143
Alvin Chin – Social Intelligence and Networking workshop, IEEE SocialCom’09
Modified SCAN method: 1. Select• Find only potential members and remove the
rest that do not meet criteria (cutoff) • Possible selection criteria
• Network centrality (Frievolt and Bielikova, 2005; Newman and Girvan, 2004)
• Subnet density (Herring et al, 2004)• Indegree/outdegree (Kumar et al, 1999; Ali-
Hasan and Adamic, 2007)
Alvin Chin – Social Intelligence and Networking workshop, IEEE SocialCom’09
Network centrality
Betweennesscentrality
Degree centrality Closeness centrality
6
8
12
6
2010
11
610
17 15
12
AB
C
DE
G H
8
15
16
F
Alvin Chin – Social Intelligence and Networking workshop, IEEE SocialCom’09
Determining cutoff centrality
Alvin Chin – Social Intelligence and Networking workshop, IEEE SocialCom’09
Modification to original SCAN Select
• Use cutoff betweenness centrality, then cutoff degree and closeness centrality• Associated with strong sense of community
(Chin and Chignell, 2007)• All members whose centrality < cutoff are
removed
Alvin Chin – Social Intelligence and Networking workshop, IEEE SocialCom’09
Modified SCAN method: 2. Collect• Find subgroups from interactions amongst
potential members found in Select• Use weighted average hierarchical clustering
• Computationally efficient than k-plex analysis (Chin and Chignell, 2007)
• Output is dendrogram – set of nested, non-overlapping clusters
Alvin Chin – Social Intelligence and Networking workshop, IEEE SocialCom’09
Weighted average hierarchical clustering
Alvin Chin – Social Intelligence and Networking workshop, IEEE SocialCom’09
Agenda1. Motivation2. Problem3. Solution4. Finding cohesive subgroups and important
members5. Case study: Nokia Friend View6. Results7. Conclusion and Future Work
Alvin Chin – Social Intelligence and Networking workshop, IEEE SocialCom’09
Nokia Friend View on the web (friendview.nokia.com)
Alvin Chin – Social Intelligence and Networking workshop, IEEE SocialCom’09
Nokia Friend View on mobile (friendview.nokia.com)
Map Status updates from friends and me
Comments to status update
Alvin Chin – Social Intelligence and Networking workshop, IEEE SocialCom’09
Finding subgroups and relevant members in Friend View• Does the friend network influence the
interaction network? Are the connections in the interaction network representative of the friend connections in the friend network?
• Do subgroup members send more comments than others in the interaction network?
• Do subgroup members have more friends than others in the interaction network?
Alvin Chin – Social Intelligence and Networking workshop, IEEE SocialCom’09
Agenda1. Motivation2. Problem3. Solution4. Finding cohesive subgroups and important
members5. Case study: Nokia Friend View6. Results7. Conclusion and Future Work
Alvin Chin – Social Intelligence and Networking workshop, IEEE SocialCom’09
Relevant members in friend network – top 10
Alvin Chin – Social Intelligence and Networking workshop, IEEE SocialCom’09
Subgroups in friend network – top 10
Alvin Chin – Social Intelligence and Networking workshop, IEEE SocialCom’09
Relevant members in interaction network –top 10
Alvin Chin – Social Intelligence and Networking workshop, IEEE SocialCom’09
Subgroups in interaction network – top 10
Alvin Chin – Social Intelligence and Networking workshop, IEEE SocialCom’09
Does the friend network influence the interaction network?
• 9 out of top 10 in friend network are in interaction network• 5 out of top 10 in friend network are in top 10 interaction
network
Yes!
Alvin Chin – Social Intelligence and Networking workshop, IEEE SocialCom’09
Do subgroup members send more comments and have more friends than others in the interaction network?
Subgroup members Interaction networkFeature
Avg # of comments 212.9 7.775
Avg # of friends 57.8 3.075
• Members that are friends of each other, are clustered together from their conversations to form subgroups
• Members of cohesive subgroups interact with each other significantly more than non-subgroup members and are most likely to be friends with each other
Yes!
Alvin Chin – Social Intelligence and Networking workshop, IEEE SocialCom’09
Conclusions
• Described modified SCAN method for finding cohesive subgroups and relevant members
• Applied method to Nokia Friend View• Discovered
• Top 10 members in the interaction network are friends of each other
• Subgroup members have more comments and friends
Alvin Chin – Social Intelligence and Networking workshop, IEEE SocialCom’09
Future work
• Content analysis of messages in Friend View• Analyze the Friend View dataset in different
time periods• Based on previous work: TorCamp study
(Chin and Chignell, 2008)• Create a recommendation and ranking
algorithm based on our method
Alvin Chin – Social Intelligence and Networking workshop, IEEE SocialCom’09
friendview.nokia.com
Alvin Chin – Social Intelligence and Networking workshop, IEEE SocialCom’09
Alvin Chinhttp://research.nokia.com/people/alvin_chin
http://www.alvinychin.com/blog
Friend View: gadgetman
Facebook: Alvin Chin
LinkedIn: [email protected]
Twitter: gadgetman4u
FriendFeed: gadgetman4u
Google: [email protected]