@article{oai:ipsj.ixsq.nii.ac.jp:00010662, author = {高木, 徹 and 藤井, 敦 and 石川, 徹也 and Toru, Takaki and Atsushi, Fujii and Tetsuya, Ishikawa}, issue = {4}, journal = {情報処理学会論文誌}, month = {Apr}, note = {類似文書検索において,検索質問文書に含まれる複数の主題を用いた高精度検索方式を提案する.本方式は,検索質問から抽出した主題ごとに通常の類似文書検索を行い,主題ごとに算出する主題重要度を用いて,最終的な類似文書検索結果を生成する.各主題から抽出される検索語の特定性をエントロピーを用いて算出することにより,各主題重要度を決定する.本方式を特許の請求項を検索質問とする無効特許検索に応用する.特許文書での請求項の前提部分や本質部分といった記述形式や構造情報を用いて,各主題重要度の補正を行う.NTCIR特許文書テストコレクションを用いた評価実験により,提案手法が従来手法より高精度な検索が可能であることを示す., In this paper, we propose an associative document retrieval method by query subtopic analysis. Our method uses the individual subtopic elements in a query document and retrieves the associative documents on a subtopic-by-subtopic basis. For each subtopic element, a subquery is produced and similar documents are retrieved with the relevance score. The relevance scores weighed by the importance of each subtopic element are integrated to determine the final relevant documents. In calculation of subtopic's importance, the specificity of a query term is evaluated using entropy, which is a deviation degree of term occurrences in each subtopic element. We applied the proposed method to an invalidity patent search in which subtopics are the composition elements in a query claim. We propose an additional calculation method of subtopic's importance using the feature of query patent claim, such as preamble and essential portions. We evaluated our method experimentally using the NTCIR patent IR test collection. The results showed that our method was effective than existing methods in retrieval accuracy.}, pages = {1074--1081}, title = {検索質問の主題分析に基づく類似文書検索と特許検索への応用}, volume = {46}, year = {2005} }