Using link and content to detect social communities

In social network analysis, community detection is an important task that aims at uncovering hidden community structure. Most of the existing methods only consider link structure in networks. However, many of them are affected by detectability threshold, a limitation that may leads to ill-defined communities. Moreover, there is link noise in networks, which makes the task more challenging. Fortunately, vertices are often associated with textual content, which is a reasonable complement for identifying good partitions. In this work, we propose an algorithm CLICT to detect social communities. The work consists of three steps: 1) expansion of social network with content similarity; 2) initial partition for weighted network; 3) refinement by triangle participation ratio. Experimental results on two real social networks demonstrate that the proposed algorithm is effective for community detection.