Presentation 2007-05-25
Text Classification Method Based on Hierarchical Relations among Words
Katsuji BESSHO, Toshio UCHIYAMA, Ryoji KATAOKA,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) Most of conventional text classification methods use symmetric measures such as cosine similarity or Euclidean metric of feature vectors derived from documents. This paper introduces to use Kullback-Leibler metric, which is non-symmetric, of the vectors for the text classification. We found that Kullback-Leibler metric instead of cosine similarity can reveal different aspect of association between words. Considering this property, this paper proposes the combined classification method which uses both cosine-Euclidean measure and KL metric. The experimental results showed that the proposed method improves classification accuracies of the method using only the cosine-Euclidean measure.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) Concept Vector / Kullback-Leibler Metric / Text Classification
Paper # PRMU2007-15,MI2007-15
Date of Issue

Conference Information
Committee PRMU
Conference Date 2007/5/17(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 Pattern Recognition and Media Understanding (PRMU)
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) Text Classification Method Based on Hierarchical Relations among Words
Sub Title (in English)
Keyword(1) Concept Vector
Keyword(2) Kullback-Leibler Metric
Keyword(3) Text Classification
1st Author's Name Katsuji BESSHO
1st Author's Affiliation NTT Cyber Solutions Laboratories, NTT Corporation()
2nd Author's Name Toshio UCHIYAMA
2nd Author's Affiliation NTT Cyber Solutions Laboratories, NTT Corporation
3rd Author's Name Ryoji KATAOKA
3rd Author's Affiliation NTT Cyber Solutions Laboratories, NTT Corporation
Date 2007-05-25
Paper # PRMU2007-15,MI2007-15
Volume (vol) vol.107
Number (no) 57
Page pp.pp.-
#Pages 6
Date of Issue