使用者能透過資訊檢索(Information Retrieval)系統快速取得資訊，因此如何提升檢索系統之效能則相當值得探討。其中，許多學者以相關回饋(Relevance Feedback)之方式改善檢索系統之效能，而在大多數的查詢擴展(Query expansion)之研究中，皆以字詞(Term)之角度進行研究，而較少考量各字詞間之語意關係。此外，有學者利用負相關回饋之資訊建立負向字典檔，將檢索結果文件中包含於負向字典檔之字詞給予字詞敏感度調整其權重，以文件重排序(document re-ranking)，並以實驗驗證其效能。然而該研究中僅針對負向字典檔計算字詞敏感度，未深入探討各字詞之資訊。因此，本研究以主題萃取之方式，分析字詞資訊與各字詞間之語意關係，擷取較具代表使用者資訊需求(information need)之主題字詞，並深入分析各主題字詞之出現情況，給予適當之字詞敏感度，以提升檢索系統之效能，進而達到檢索結果更符合使用者之資訊需求。並以實驗顯示本研究提出之演算法能提升檢索之效能，且相較於傳統之Rocchio演算法有較佳之表現。;Users can get information by using information retrieval systems, so how to improve performance of systems is worthwhile discussion. Many researchers used relevance feedback to improve performance of systems. And in most of query expansion studies, they focused on term, but seldom on semantic relationships. In past study, the researcher applied the information of non-relevant documents to create a negative dictionary. Then, used the features of this negative dictionary to adjust term weights of retrieved documents to re-rank documents. The researcher just focused on the information of non-relevant documents, but didn’t use other information of words. So, this study improve the efficiency of information retrieval and make it more suitable for users by using their relevance feedback, and the new word sensitivity, which is built by using concept-retrieval algorithm and analyzing term-appearance situation. The result of experiments show that the proposed method of this study is effectiveness in document retrieval.
