User interaction-oriented community detection based on cascading analysis

Detecting user communities in online social networks (OSNs) is of great importance for understanding social dynamics. Compared to user connections, user interactions have been shown to be more meaningful for reflecting peer relationship in OSNs. To this end, we propose a user interaction-oriented community detection method based on cascading analysis. Specifically, user interactions are analyzed from a large collection of social object sharings (e.g., blog posts, photo shares). Both direct and indirect user interactions associated with each social object sharing are then extracted and the cascading relations among these interactions are captured using a graph representation. The proposed method makes use of such cascading relations to extract groups of actively interacting users and adopts a super graph approach to cluster these user groups for detecting communities. An extensive evaluation of our method was performed using three real OSN datasets and compared with three state-of-the-art overlapping community detection methods, namely two general methods applied to the interaction graph and an interaction-based method. Our method outperformed the compared methods, as demonstrated by several evaluation metrics, and produced more robust and stable detection results across different datasets.