在資訊科技蓬勃發展的時代，資訊檢索 (Information Retrieval)系統對人們而言是獲取資訊的重要管道之一。如何使資訊檢索系統更貼近使用者的資訊需求，一直以來都是被高度關注的研究議題。在相關回饋(Relevance Feedback)領域中，以Rocchio演算法最為廣泛使用，此演算法透過字詞出現的頻率來分析出相關回饋結果中的重要字詞，作為查詢擴展(Query Expansion)的字詞來源，其實作結果具有一定水平的效能且經常被應用於各種檢索中。不過，Rocchio僅以字詞的出現頻率為相關文件的重要字詞依據，未考慮到字詞間是否有其他特性可以運用。本研究運用相關回饋的結果做語詞關係分析，以找出適合作為查詢擴展的字詞，並透過負相關回饋之分析結果修正正相關回饋之分析結果，藉此運用字詞間不同的特性，改善只考慮字詞頻率而忽略其他語詞關係的方法，進而提升文件檢索之準確率。透過實驗證明本研究所提出之方法相較於其他方法，有較佳的檢索效果。;With the quick development of the information technology, information retrieval system is one of the critical way for people to get information. The research of how the result of information retrieval system can get closer to users’ information need has always been highly concerned. Rocchio’s query expansion algorithm has been widely utilized in relevance feedback. It analyzed the importance of terms residing in relevance feedback by term frequency, and be the source of query expansion terms. The algorithm has been applied in different retrieval due to its simple and effectiveness. However, Rocchio algorithm only focuses on term frequency and ignores other terms’ relation that may be useful. Therefore, in this research, we aimed to develop a method applies terms’ relation in relevance feedback, including synonymy and co-occurrence relation, to analyze terms which are suitable for query expansion. Then, revising relevance feedback query result by non-relevance result to do document re-ranking. Through applying different relation between terms in relevance and non-relevance documents can improve the method only focuses on term frequency and ignores others. The results of experiments verify that the proposed method is effectiveness in document retrieval.
