DSpace university logo mark
Advanced Search
Japanese | English 

NAOSITE : Nagasaki University's Academic Output SITE > Faculty of Engineering > Conference Paper >

A Revised Inference for Correlated Topic Model


File Description SizeFormat
LNCS7952_445.pdf1.74 MBAdobe PDFView/Open

Title: A Revised Inference for Correlated Topic Model
Authors: Masada, Tomonari / Takasu, Atsuhiro
Issue Date: Jul-2013
Publisher: Springer Verlag
Citation: Lecture Notes in Computer Science, 7952, pp.445-454; 2013
Abstract: In this paper, we provide a revised inference for correlated topic model (CTM) [3]. CTM is proposed by Blei et al. for modeling correlations among latent topics more expressively than latent Dirichlet allocation (LDA) [2] and has been attracting attention of researchers. However, we have found that the variational inference of the original paper is unstable due to almost-singularity of the covariance matrix when the number of topics is large. This means that we may be reluctant to use CTM for analyzing a large document set, which may cover a rich diversity of topics. Therefore, we revise the inference and improve its quality. First, we modify the formula for updating the covariance matrix in a manner that enables us to recover the original inference by adjusting a parameter. Second, we regularize posterior parameters for reducing a side effect caused by the formula modification. While our method is based on a heuristic intuition, an experiment conducted on large document sets showed that it worked effectively in terms of perplexity.
Description: 10th International Symposium on Neural Networks, ISNN 2013; Dalian; China; 4 July 2013 through 6 July 2013
Keywords: covariance matrix / Topic models / variational inference
URI: http://hdl.handle.net/10069/33725
ISSN: 03029743
DOI: 10.1007/978-3-642-39068-5_54
Rights: © 2013 Springer-Verlag Berlin Heidelberg.
Type: Conference Paper
Text Version: author
Appears in Collections:Conference Paper

Citable URI : http://hdl.handle.net/10069/33725

All items in NAOSITE are protected by copyright, with all rights reserved.

 

Valid XHTML 1.0! Copyright © 2006-2015 Nagasaki University Library - Feedback Powerd by DSpace