GroupTracker: Incremental Extraction of Fuzzy Groups from Over-Time Network Data
Larry Richard Carley, Kathleen CarleyThis talk will describe GroupTracker, a novel efficient algorithm for extracting fuzzy group structures from over-time social media data sets. The method can efficiently extract fuzzy groups, which are group structures in which individual actors can be in multiple groups with different weights, at each point in time. The algorithm calculates the weights for each individual actor being in each possible group structure at each point in time; however, this requires knowing the number of groups at each point in time. Therefore, GroupTracker also employs an efficient incremental algorithm for determining the “best” number of groups at any point in time in an over-time social media data set.
Both the determination of the “best” number of groups and then the determination of the weights of each individual in each group at each point in time require the selection of a scalar measure for the goodness of a grouping result. Girvan and Newman defined the “Modularity” of binary symmetric networks as one measure for the goodness of a grouping result and they used it to identify group structures by deleting links with high betweenness centrality. However, real world networks of interest are often weighted and not symmetric. In addition, it is often the case that individuals are members of more than one group. Therefore, GroupTracker is designed to allow the user to select from several options for the scalar measure of group quality. Results of preliminary trials using several measures for group quality applied to several real-world over-time social media data sets will be presented and discussed.