Group and Topic Discovery from Relations and Their Attributes

摘要

The authors present a probabilistic generative model of entity relationships and their attributes that simultaneously discovers groups among the entities and topics among the corresponding textual attributes. Block- models of relationship data have been studied in social network analysis for some time. Here, the authors simultaneously cluster in several modalities at once, incorporating the attributes (here, words) associated with certain relationships. Significantly, joint inference allows the discovery of topics to be guided by the emerging groups, and vice-versa. They present experimental results on two large data sets: 16 years of bills put before the U.S. Senate, including their corresponding text and voting records, and 13 years of similar data from the United Nations. The authors show that in comparison with traditional, separate, latent-variable models for words, or Block-structures for votes, the Group-Topic model's joint inference discovers more cohesive groups and improved topics.