Paper 3

COTILES: Leveraging Content and Structure for Evolutionary Community Detection

Authors: Nikolaos Sachpenderis, Georgia Koloniari, Alexandros Karakasidis

Volume 45 (2020)

Abstract

Most community detection algorithms for online social networks rely solely either on the structure of the network, or on its contents. Both extremes ignore valuable information that influences cluster formation. We propose COTILES, an evolutionary community detection algorithm, that leverages both structural and content-based criteria so as to derive densely connected communities with similar contents. Specifically, we extend a fast online structural community detection algorithm by applying additional content-based constraints. We also further explore the effect of structure and content-based criteria on the clustering result by introducing three tunable variations of COTILES that either tighten or relax these criteria. Through our experimental evaluation, we show that the proposed method derives more cohesive communities compared to the original structural one, and highlight when the proposed variations should be deployed.

Keywords

Community detection, Social networks, Labeled communities, Evolutionary clustering