Entity Role Discovery in Hierarchical Topical Communities

People and social communities are often characterized by the topics and themes they are working on, or communicating about. Discovering the roles played by dierent entities in these communities are of great interest in many real-world contexts in social network analysis. We are also often interested in discovering such roles at dierent levels of granularity. In this paper we study a new problem of mining entity roles in hierarchical topical communities. We rst detect topical communities from the text component of a social or information network. Since we mine phrases from the network, and represent topical communities by ranked lists of mixed-length phrases, the communities have a good interpretation at multiple levels of the hierarchy. We are therefore able to discover topical roles of dierent types of entities in both large communities encompassing more general topics, and small, focused subcommunities. We demonstrate our method on a bibliographic information network dataset, which we use to discover the roles of authors and publication venues in the context of the hierarchical topical communities.