Microgroup mining on TSina via network structure and user attribute

In this paper, we focus on the problem of community detection on TSina: the most popular microblogging network in China. By characterizing the structure and content of microgroup (community) on TSina in detail, we reveal that different from ordinary social networks, the degree assortativity coefficients are negative on most microgroups. In addition, we find that users from the same microgroup likely exhibit some similar attributes (e.g., sharing many followers, tags and topics). Inspired by these new findings, we propose a united method for microgroup detection without losing the information of link structure and user attribute. First, the link direction is converted to the weight by giving higher value to the more surprising link, while attribute similarity between two users is measured by the Jaccard coefficient of common features like followers, tags, and topics. Then, above two factors are uniformly converted to the edge weight of a newly generated network. Finally, many frequently used community detection algorithms that support weighted network would be employed. Extensive experiments on real social networks show that the factors of link structure and user attribute play almost equally important roles in microgroup detection on TSina. Our newly proposed method significantly outperforms the traditional methods with average accuracy being improved by 25%, and the number of unrecognized users decreasing by about 75%.