An Empirical Study for Digest Generation : Constituent Word Correspondence between Titles and Body Parts of Japanese Articles

Bibliographic Information

Other Title
  • 見出しを利用した新聞・レポートからのダイジェスト情報の抽出

Search this article

Description

This paper presents a simple method of automatic digest generation of articles, which is intended to provide an effective view of a large amount of retrieved texts. The method generates a digest using a seed list which contains constituent nouns of title. A digest generator repeats sentence selection until the seed list is empty, in which it calculates sentence relevance based on the overlaps between the seed list and a sentence, and it picks up the most relevant sentence and eliminates its constituent nouns from the seed list. It generated one to three sentences digests from 93% of 13,562 newspaper articles and 61 of 62 economic reports. 98 samples of them are evaluated manually and more than 82% of them are judged to be enough understandable to recognize outline of the source articles.

Journal

  • IPSJ SIG Notes

    IPSJ SIG Notes 117 121-128, 1997-01-20

    Information Processing Society of Japan (IPSJ)

Citations (12)*help

See more

References(7)*help

See more

Details 詳細情報について

  • CRID
    1571698602153242240
  • NII Article ID
    110002934640
  • NII Book ID
    AN10115061
  • Text Lang
    ja
  • Data Source
    • CiNii Articles

Report a problem

Back to top