本文介紹我們以向量空間檢索模式、N-gram索引法及關鍵詞自動擷取技術開發的 「模糊搜尋」、「相關詞提示」與「相關詞回饋」的檢索功能， 並評估其整合在既有 OPAC 系統上的應用成效。我們採用線上問卷調查法、檢索記錄檔分析法及資訊檢索領域常用的查 全率、查準率等指標來衡量使用者的反應及檢索成效。主要的結果如下：一、新的檢索功能 使用比例高，且此情形與受訪者對這些功能的熟悉程度無關，但與受訪者無法精確表達查詢 條件有關；二、多數受訪者能遵循相關詞提示，繼而模糊搜尋，之後繼續採用模糊搜尋或相 關詞回饋的最有利過程進行資料查詢，而且此現象也與受訪者對這些功能的熟悉度無關；三 、不論受訪者是否能夠精確表達查詢條件，新的檢索功能沒有比傳統欄位式檢索系統提供更 佳的檢索結果滿意度，惟在找到資料的情形上表現較好； 四、多數受訪者（ 80% ）對於系 統自動擷取的詞彙，偶有不合理的情形（ 10% ）多可接受； 五、在理想的檢索環境下，相 關詞提示與相關詞回饋比用模糊搜尋直接查詢資料庫的成效更佳， 分別可提升 38.2 ％及 29.1 ％的檢索成效。

　　　　　This paper introduces three retrieval functions based on the vector space retrieval model, N-gram indexing model, and an automatic keyword extraction technique. These retrieval functions are known as fuzzy search, term suggestion, and term relevance feedback. Evaluation of these new functions integrated in an OPAC system is performed through the approaches of online questionnaires, search log analysis, and recall and precision rates. The main results: (1) The new retrieval functions have high usage rates, which is independent of the users' familiarity with the functions, but is related to the inability to express queries exactly. (2) Most users follow the best retrieval steps by first using term suggestion, and then fuzzy search, and then continuing on fuzzy search or changing to term relevance feedback, which is also independent of the users' familiarity with these functions. (3) No matter whether users can express their queries exactly or not, new retrieval functions do not provide better retrieval satisfaction than traditional Boolean search. However, they do perform better in providing better results. (4) The 10% illegal suggested terms, automatically extracted by the system, are acceptable by 80% users. (5) In ideal environments, term suggestion and term relevance feedback perform better than fuzzy search, with performance increases of 38.2% and 29.1%, respectively.