網際網路搜尋引擎為現今快速獲取資訊的重要工具，但其所搜尋到的網頁不僅在量上過於龐大，更常是與使用者需求不符之結果。因此，個人化搜尋的需求也因應而生。本研究提出了一套演算法，用於萃取網頁文件特徵，希望透過文件特徵間的相似度比對，分辨出搜尋結果中與使用者需求相關、以及非相關的文件，並藉此過濾掉非相關文件。其中，特徵權重值計算的部份包含了四項因子：HTML強調標籤權重、段落權重、分析組合權重、以及字詞敏感度權重。我們分析各因子對文件特徵的影響，並且與相關演算法作比較，以了解本研究演算法之優劣所在。實驗結果證實，本研究演算法能夠有效萃取出網頁文件中的重要特徵。利用上述文件特徵所建置之使用者興趣檔，能夠提升與相關文件的相似度，及降低與非相關文件的相似度，藉此有效過濾與使用者需求不相關之網頁文件搜尋結果。 Nowadays, Web search engine has become an important tool to get information rapidly. However, there are too many searching results retrieved from search engine, and always, these searching results do not conform to user's request. To reduce personal effort on information searching, personal search are required. In this research, we present a method to extract Web document feature, and by way of comparing the similarities of document features, we could better recognize which documents are conformed to user's request. The method of document feature extraction includes four factors: HTML emphasis tag, term position, analytic combination of criteria, and term sensitivity. The results of our experiment show that our method can extract important features of Web document efficiently. The user profile consists of the above document features could increase the similarity with relevant documents, and decrease the similarity with irrelevant documents.
