Computer Science Colloquium
Thursday, November 14, 4:15pm, room 9205/06
Bayesian nonparametric tree-structured topic models and big data
We discuss a tree-structured topic model that uses Bayesian nonparametrics to learn the branching factor of each node. Our approach models each document as a distribution over a subtree of topics, where topics become more specific as they move away from the root node. We discuss an efficient inference algorithm for model learning when the number of documents is in the millions.
(there will be no wine and cheese reception)
The Colloquium is supported by generous contributions from the Bloomberg, Information Builders, Inc., and Netlogic, Inc.