TheAuthor-TopicModelforAuthorsandDocuments-Abstract.PDFVIP

In 20th Conference on Uncertainty in Artificial Intelligence. Banff, Canada The Author-Topic Model for Authors and Documents Michal Rosen-Zvi Thomas Gri?ths Mark Steyvers Padhraic Smyth Dept. of Computer Science Dept. of Psychology Dept. of Cognitive Sciences Dept. of Computer Science UC Irvine Stanford University UC Irvine UC Irvine Irvine, CA 92697-3425, USA Stanford, CA 94305, USA Irvine, CA 92697, USA Irvine, CA 92697-3425, USA Abstract Here, we consider how these approaches can be used to address another fundamental problem raised by large We introduce the author-topic model, a gen- document collections: modeling the interests of au- erative model for documents that extends La- thors. tent Dirichlet Allocation (LDA; Blei, Ng, By modeling the interests of authors, we can answer Jordan, 2003) to include authorship informa- a range of important queries about the content of tion. Each author is associated with a multi- document collections. With an appropriate author nomial distribution over topics and each topic model, we can establish which subjects an author is associated with a multinomial distribution writes about, which authors are likely to have writ- over words. A document with multiple au- ten documents similar to an observed document, and thors is modeled as a distribution over topics which authors produce similar work. However, re- that is a mixture of the distributions associ- search on author modeling has tended to focus on the ated with the authors. We apply the model problem of authorship attribution – who wrote which to a collection of 1,700 NIPS conference pa- document – for which