Presentation 2014-05-15
Document Summarization Using Word Vectors
Katsuji BESSHO, Hitoshi NISHIKAWA, Toshiro MAKINO, Yoshihiro MATSUO,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) As a technique of document summarization, we verified a technique of expressing a word as a topic vector, and expressing a sentence and a document as a composition of the vectors of constituent words, and computing the score of a sentence based on a similarity with the vector of the subject document, and outputting a high-scored sentence as a summary text. We conducted an experiment under the constraints of inputting a document which consists of the list of text blocks, and of outputting the summary text as one sentence or one word for every topic. The results indicate that our proposed method of using a word vector achieved a higher F-score compared to the baseline technique that uses the sum or the average of a word score.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) Extraction-Based Summarization / Word Vector / Document Vector / Centroid
Paper # LOIS2014-5
Date of Issue

Conference Information
Committee LOIS
Conference Date 2014/5/8(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 Life Intelligence and Office Information Systems (LOIS)
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) Document Summarization Using Word Vectors
Sub Title (in English)
Keyword(1) Extraction-Based Summarization
Keyword(2) Word Vector
Keyword(3) Document Vector
Keyword(4) Centroid
1st Author's Name Katsuji BESSHO
1st Author's Affiliation NTT Media Intelligence Laboratories()
2nd Author's Name Hitoshi NISHIKAWA
2nd Author's Affiliation NTT Media Intelligence Laboratories
3rd Author's Name Toshiro MAKINO
3rd Author's Affiliation NTT Media Intelligence Laboratories
4th Author's Name Yoshihiro MATSUO
4th Author's Affiliation NTT Media Intelligence Laboratories
Date 2014-05-15
Paper # LOIS2014-5
Volume (vol) vol.114
Number (no) 32
Page pp.pp.-
#Pages 6
Date of Issue