Discovering leaders from social network by action cascade

This paper proposes an approach to discover community leaders in a social network by means of a probabilistic time-based graph propagation model. In the study, community leaders are defined as those people who initiate a lot of influence chains according to the number of chains they belong to. To conduct the approach, we define an exponential time-decay function to measure the influence of leaders and construct the chains of leaders' action-specific influence. Then, we build the general chains by normalizing over all possible users' actions. In specific, this paper uses the variant of the Apriori algorithm called APPM to mine users' influence paths. To the best of our knowledge, this work is the first attempt to use the action-specific influence chains for mining community leaders from a social network. In our experiments, two datasets are collected to examine the performance of the proposed algorithm: one is a social network dataset from Facebook consisting of 134 nodes and 517 edges, and another is a citation network dataset from DBLP and Google Scholar containing 2,525 authors, 3,240 papers, and 1,914 citations. In addition, several baselines are also carried out for comparison, including three naive and one user-involved approaches. The experimental results show that, compared with the baselines, for the task of detecting community leaders, the proposed method outperforms the baselines, and our method also obtains good performance on ranking community leaders.