中文問答系統－以網路為基礎之查詢詞擴充策略

Abstract

網路有如資訊的海洋。然而遺憾的是，人們在網路中尋找自己感興趣的答案常如大海撈針。傳統的關鍵詞檢索方式多無法解決使用者的查詢意圖。 本論文提出一個應用網路資料和查詢詞擴充技術的中文問答系統。我們提出法則式的問句樣式機制以分析問句的意圖。另一方面，有別於一般中文問答系統擴充詞多來自事先所設定的相關詞資料庫，本論文所提的查詢詞擴充技術乃是應用現成的網路語料，進行相關詞探勘。我們利用對應演算法將訓練問句和搜尋結果進行非名詞關鍵詞與查詢詞擴充。 爲了檢驗所提的方法，我們以383個問句做為訓練資料，進行查詢詞擴充探勘，並另以80個問句作測試，所得到的搜尋結果比一般關鍵詞搜尋在使用者所需要閱讀的篇數明顯減少，實驗結果顯示系統效能為每個問題所花的human effort 2.03 和MMR 0.765。

Searching in the Web is just like searching in a sea. Traditional query resolution is based on inefficient keyword search. This thesis proposes a Chinese question answering system by using web corpus and query expansion. We propose a rule-based query processing method to detect the query type. On the other hand, we propose new query expansion which is unlike traditional one based on predefined thesaurus. The presented query expansion is based on web corpus by aligning the training questions and the search-results returned from a search engine. In order to verify the proposed method we use 383 questions for training and 80 questions for testing. The results show that the proposed expansion technique yields better performance than the keyword-based search in terms of less human efforts per question 2.03 and MMR 0.765.