Presentation 2009-05-22
Latent Variable Model Focusing on Documents' Topics And Authors' Interests
Noriaki KAWAMAE, Takeshi YAMADA,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) This paper presents a statistical model that captures not only the low-dimensional set of multinomial distributions over words, but also how this structure is consisted of topics, document classes and author interests. Unlike other recent work that includes authors into the latent variable model, here the interest is represented as a latent variable having a probability distribution over topics and can be shared with authors who prefer a set of similar topics. So, this model represents each document as a mixture of topics, where the mixture proportion is sampled from this interest class. Experiments using a dataset of research papers shows that the proposed model can capture these interests, thus performs dimensionality reduction of documents to a low-dimensional interest-topic space, and is useful as a generative model.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) latent variable models / clustering
Paper # AI2009-4
Date of Issue

Conference Information
Committee AI
Conference Date 2009/5/15(1days)
Place (in Japanese) (See Japanese page)
Place (in English)
Topics (in Japanese) (See Japanese page)
Topics (in English)
Chair
Vice Chair
Secretary
Assistant

Paper Information
Registration To Artificial Intelligence and Knowledge-Based Processing (AI)
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) Latent Variable Model Focusing on Documents' Topics And Authors' Interests
Sub Title (in English)
Keyword(1) latent variable models
Keyword(2) clustering
1st Author's Name Noriaki KAWAMAE
1st Author's Affiliation NTT Communication Science Laboratories()
2nd Author's Name Takeshi YAMADA
2nd Author's Affiliation NTT Communication Science Laboratories
Date 2009-05-22
Paper # AI2009-4
Volume (vol) vol.109
Number (no) 51
Page pp.pp.-
#Pages 6
Date of Issue