Search

In recent times, due to the excessiveness of research paper in the web, there is a need of automatic document summarization for papers. And, text summarization is depend on keyword extraction. In this paper, we propose keyword extraction method combining linguistics and graph theory proposed in related work. By inputting the co-occurrence, position, the term frequency, and inverse document frequency of word, this method provide words feature score.

文圧縮は,文のもつ情報をできるかぎり保持したまま,より短い文を生成するタスクである.既存の文圧縮手法の多くは,原文の構文木の構成素を取り除く単純な操作のみに基づいており,構文木の構造的変更に対応することはできない.本稿では,変更操作の対象として,構文木中の再帰構造に注目し,それを取り除く文圧縮手法を提案する.構文木に出現する再帰的な構造を検出し,これを除去する操作を導入することで,文法性を保ったまま再帰構造を含む文を圧縮できる.評価実験の結果,圧縮された文の文法性について,提案手法が既存の手法に比べて優れていることを確認した., Sentence compression is a task of generating a grammatical short sentence from an original sentence, retaining the most important information. The existing methods of only removing the constituents in the parse tree of an original sentence cannot emulate human compression which changes structures of the parse tree. This paper proposes a method to remove recursive structures, one of such structural conversions, and generate a grammatical short sentence. In order to remove a recursive structure, our method detects the constituents forming the structure and remove them as a unit. Compression experiments have shown that our method generates more grammatical compressed sentences than the previous method.

Supervisor:國藤 進, 知識科学研究科, 修士