ECODE: Event-Based Community Detection from Social Networks

Xiaoli Li, Aloysius Tan, Philip S. Yu, See-Kiong Ng

Abstract: People regularly attend various social events to interact with other community members. For example, researchers attend conferences to present their work and to network with other researchers. In this paper, we propose an E vent-based COmmunity DEtection algorithm ECODE to mine the underlying community substructures of social networks from event information. Unlike conventional approaches, ECODE makes use of content similarity-based virtual links which are found to be more useful for community detection than the physical links. By performing partial computation between an event and its candidate relevant set instead of computing pair-wise similarities between all the events, ECODE is able to achieve significant computational speedup. Extensive experimental results and comparisons with other existing methods showed that our ECODE algorithm is both efficient and effective in detecting communities from social networks.