Detecting Communities Using Link and Content Triangles

Community detection for uncovering the hidden community structure in large networks is an important task in analyzing complex networks. Most of the existing methods only consider link structure in networks, where the link information is usually sparse and noisy, which may result in a poor partition of a network. Fortunately, besides link structure, nodes, especially in social networks, are often associated with certain symbolic or textual attributes, which we refer to as content. Content, therefore, is expected to serve as a reasonable complement for finding a good partition. In this work, we propose an algorithm LICT to detect communities with link and content triangles. It works in three steps: 1) network expansion with content similarity; 2) community detection in weighted network; and 3) refinement by weighted triangle modularity. Experimental results on several real data sets demonstrate that the proposed algorithm is effective for community detection and robust in the presence of link noise.