PASCAL - Pattern Analysis, Statistical Modelling and Computational Learning

Fast Inference in Conditional Topic Models
Philipp Hennig, David Stern, Thore Graepel and Ralf Herbrich
In: ICML 2011, Jun 28 - Jul 2 2011, Bellevue, Washington, USA.

Abstract

Topic models use word frequencies to describe semantic similarity between documents in a low-dimensional latent space. Modern document repositories often record metadata in addition to the words themselves, which can convey important semantic information. Because such corpora can also be very large, inference should be computationally lightweight. We construct a fast approximate inference scheme for topic models conditional on arbitrary features of the document. We also study the viability of single pass inference in such models, and show experimental results from large online document corpora. The result is the first “web-scale” conditional topic model.

PDF - Archive staff only - Requires Adobe Acrobat Reader or other PDF viewer.
EPrint Type:Conference or Workshop Item (Paper)
Project Keyword:Project Keyword UNSPECIFIED
Subjects:Computational, Information-Theoretic Learning with Statistics
Learning/Statistics & Optimisation
Natural Language Processing
Theory & Algorithms
Information Retrieval & Textual Information Access
ID Code:7208
Deposited By:Philipp Hennig
Deposited On:09 March 2011