抄録からの主題文の自動抽出

原田隆史; Takashi Harada; 丸山宏; Hiroshi Maruyama; 佐藤元美; Motomi Satoh; 細野公男; Kimio Hosono; 諸橋正幸; Masayuki Morohashi

doi:10.46895/lis.29.125

抄録からの主題文の自動抽出Automatic selection of the subject sentences from abstracts bearing

原田隆史¹，丸山宏²，佐藤元美³，細野公男⁴，諸橋正幸²Takashi Harada¹, Hiroshi Maruyama², Motomi Satoh³, Kimio Hosono¹, Masayuki Morohashi²

¹ 慶鷹義塾大学文学部図書館・情報学科助手Teaching Assistant, School of Library and Information Science, Keio University ◇ 〒108-8345 東京都港区三田二丁目15番45号 ◇ Mita 2-15-45, Minato-ku, Tokyo 108-8345, Japan

² 日本IBM東京基礎研究所Researcher, Tokyo Research Laboratory, IBM Japan Ltd. ◇ 東京都千代田区三番町 ◇ Sanbancho, Chiyoda-ku, Tokyo, Japan

³ 東京海上火災The Tokio Marine and Fire Insurarice Co. Ltd. ◇ 〒100-8050 東京都千代田区丸の内一丁目2番1号 ◇ Marunouchi 1-2-1, Chiyoda-ku, Tokyo 100-8050, Japan

⁴ 慶鷹義塾大学文学部図書館・情報学科教授Professor, School of Library and Information Science, Keio University ◇ 〒108-8345 東京都港区三田二丁目15番45号 ◇ Mita 2-15-45, Minato-ku, Tokyo 108-8345, Japan

受付日：1992年2月28日Received: February 28, 1992

発行日：1992年3月31日Published: March 31, 1992

In order to automatically extract good free keywords for content designation from any kind of text, it must be effective and efficient to single out subject bearing sentences at first, and then extract those keywords from them.

This paper, first of all, describes characteristics of a method developed to automatically discriminate such sentences from those which show premises and conclusions, based on the relative position of the sentences in texts and the paticular expressions appeared in them.

Then this paper reports that the result of the experiment where the method was applied to the abstracts in the field of computer science. In 286 out of 386 abstracts (69.4%), almost all sentences include in them were classified successfully. Furtheremore, some of the sentences were recognized conectly in 65 abstracts. As far as the first sentence is a subject bearing one, it was identified so with very high probability i.e. 97.1%.