自動查詢擴展技術在許多資訊檢索相關研究中，已被證實可以有效增進檢索效能，此技術主要解決檢索過程中字彙不匹配 (Word mismatch)的問題，即使用者所提供的查詢字彙與文件使用字彙間差異所導致的檢索效能不佳問題。 本研究延伸機率模型之非隨機性概念，提出一查詢擴展演算法，並使用初始檢索排名前幾篇之文件為擴展關鍵字來源，利用Rocchio架構重新衡量候選關鍵字權重與選取擴展關鍵字，再加以進行查詢擴展。 經使用單一主題領域之Cranfield與多主題領域之npl測試資料集進行實驗與分析後，結果顯示本研究提出之方法可有效提升檢索效能。除此之外，本研究也針對影響查詢擴展效能的相關參數，進行詳細的實驗與分析，這些參數包括虛擬相關文件集合篇數、擴展關鍵字數目與Rocchio架構下之參數。 Automatic query expansion addresses the problem of word mismatching that the words provided by the users in the query are not consistent with the words used by the authors. The problem of word mismatching can result in poor retrieval effectiveness. Many techniques of automatic query expansion have been developed and proved to improve retrieval effectiveness. We apply the concept of the non-randomness of probabilistic model to conceive a method for automatic query expansion. Top-ranked documents that are retrieved in the initial retrieval are used as the source of expansion terms. The candidate expansion terms are re-weighted and selected within Rocchio framework. Experimenting results show that our approach can improve the effectiveness of retrieving significantly. The experiments have the parameters that can influence the performance of automatic query expansion considered and analyzed, including number of selected documents, number of expansion terms and parameters in the Rocchio framework.
