過去在向量空間模型相關回饋研究上，多數是使用者對於查詢結果中的相關文件清單，擷取其字頻資訊為回饋結果，又以Rocchio查詢擴張最被廣泛使用，此演算法具有一定水平與效能且經常被應用在各種檢索中，但是否有其他更好判斷相關字詞和過濾不相關字詞方法，一直以來也都是個被高度關注的研究議題。近幾年語意檢索概念蓬勃發展，許多搜尋引擎開始嘗試以使用者查詢和相關文件中字詞語意做為檢索依據，這種機制主要著重在關鍵字上所涵蓋的語意概念，而非單純只考慮關鍵字本身。因此本研究基於現有運用在資訊檢索相關研究中的語意技術為基礎，從使用者提供之相關文件資訊中排除不含語意字詞並擷取具有高度語意關係字詞為查詢擴張字詞來源，試圖找出查詢和相關文件字詞間可利用的語意關係，並透過實驗證明語意應用方法可以和Rocchio演算法達到不分軒輊，甚至於更佳的效果。;In previous vector space model and relevance feedback studies, the result is determined by the frequency of the terms in the list of relevant documents after users input their queries into the information retrieval system. One of the most popular methods is Rocchio′s query expansion algorithm, which has good performance and is often used in many cases. However, it has been highly focused in researches whether there are better methods to choosing relevant words and filtering out irrelevant words. In recent years, the semantic analysis techniques are developed, which emphasizes the meaning of terms rather than just the terms themselves. Many search engines have adopted these techniques to judge the result of relevant feedback. In this research, we try to find the usable meanings between the query and relevant words. The performance of the methods based on semantic analysis techniques in this research has been evaluated in experiments and proved to work as good as Rocchio’s algorithm, in some cases even better.
