Markov-network based latent link analysis for community detection in social behavioral interactions

How to represent and discover social links from the perspective of implied behaviors, in particular latent links, is critical for social media analysis. In this paper, we discuss latent link analysis for community detection in social behavioral interactions. We adopt Markov network (MN) as the framework and propose the algorithm to discover latent links among social objects implied in their behavioral interactions without regard for the topological structures of social networks. First, starting from the frequent itemsets of the behavioral interactions, we propose the algorithm to construct the item-association Markov network (IAMN), which establishes the inherent relationship between frequent itemset and MN. Then, we propose the algorithm to detect communities by incorporating the concepts of k-clique and k-nearest neighbor set, as the typical application of the constructed IAMN Experimental results show the effectiveness and efficiency of the method proposed in this paper.