Group formation in large social networks: membership, growth, and evolution

    Proceedings of the 19th ACM SIGKDD international conference on Knowledge discovery and data mining, 2006.

    Cited by: 1949|Bibtex|Views110|Links
    EI
    Keywords:
    structural featuresocial networking sitecommunity membershiplarge-scale time-resolved datasocial groupMore(10+)
    Wei bo:
    We have considered the ways in which communities in social networks grow over time — both at the level of individuals and their decisions to join communities, and at a more global level, in which a community can evolve in both membership and content

    Abstract:

    The processes by which communities come together, attract new members, and develop over time is a central research issue in the social sciences - political movements, professional organizations, and religious denominations all provide fundamental examples of such communities. In the digital domain, on-line groups are becoming increasingly...More

    Code:

    Data:

    Summary
    • The tendency of people to come together and form groups is inherent in the structure of society; and the ways in which such groups take shape and evolve over time is a theme that runs through large parts of social science research [9].
    • The questions we consider are closely related to the diffusion of innovations, a broad area of study in the social sciences [31, 33, 34]; the particular property that is “diffusing” in our work is membership in a given group.
    • Before turning to our studies of the processes by which individuals join communities in a social network, we provide some details on the two sources of data, LiveJournal and DBLP.
    • In Figures 1 and 2 we show this basic relationship for LJ and DBLP respectively: the proportion P (k) of people who join a community as a function of the number k of their friends who are already members.
    • While these curves represent a good start towards membership prediction, they estimate the probability of joining a community based on just a single feature — the number of friends an individual has in the community.
    • By applying decision-tree techniques to these features we find that we can make significant advances in estimating the probability of an individual joining a community.
    • Both tables contain the baseline performance one could achieve by predicting based solely on the number of friends a fringe member already has in the community.
    • Given that movement bursts intuitively represent increased participation from some other community, these differences will provide a first perspective on the general question of whether topics are following people, or whether people are following topics.
    • (While this number is a useful benchmark for relative comparisons, its actual magnitude can clearly be affected by altering the settings of the burst detection parameters.) On the other hand, as shown in Table 5, 43.91% of all papers contributing to movement bursts use hot terms.
    • We have considered the ways in which communities in social networks grow over time — both at the level of individuals and their decisions to join communities, and at a more global level, in which a community can evolve in both membership and content.
    • It will be interesting to connect standard theoretical models of diffusion in social networks to the kinds of data on community membership that one can measure in on-line systems such as LiveJournal.
    • Such representations can clearly form the basis for alternate ways of quantifying community movement, with conferences forming natural groupings by topic, and with certain parts of the space becoming “filled out” as particular areas emerge over time.
    • Even with very rich data, it is challenging to formulate the basic questions here, and we view the elaboration of further questions to be an interesting direction for future work
    Funding
    • Over all communities, the mean growth rate was 18.6%, while the median growth rate was 12.7%. We cast this problem directly as a binary classification problem in which class 0 consists of communities which grew by less than 9%, while class 1 consists of communities which grew by more than 18%
    • As shown in Table 5, we find that papers contributing to movement bursts in fact use expired hot terms at a significantly higher rate than arbitrary papers at the same conference (31.02% vs. 26.37%), but use future hot terms at a significantly lower rate (11.53% vs. 17.40%)
    • In other words, of the four patterns, shared interest is 50% more frequent than the other three patterns combined
    Your rating :
    0

     

    Tags
    Comments