EgoClustering: Overlapping Community Detection via Merged Friendship-Groups

Abstract

There has been considerable interest in identifying communities within large collections of social networking data. Existing algorithms will classify an actor (node) into a single group, ignoring the fact that in real-world situations people tend to belong concurrently to multiple (overlapping) groups. Our work focuses on the ability to find overlapping communities. We use egonets to form friendship-groups. A friendship-group is a localized community as seen from an individual’s perspective that allows an actor to belong to multiple communities. Our algorithm finds overlapping communities and identifies key members that bind communities together. Additionally, we will highlight the parallel feature of the algorithm as a means of improving runtime performance, and the ability of the algorithm to run within a database and not be constrained by system memory.