搜尋引擎已成為現代人尋找資料最有力的幫手，使用者往往經由輸入查詢來獲得搜尋引擎的回傳結果，但研究顯示77%的使用者在使用搜尋引擎時，只有使用一個英文單字當作查詢，且85%的使用者所輸入的查詢字串少於三個英文單字，由此可知絕大多數的使用者在使用搜尋引擎查詢資料時，其所輸入的查詢皆為不完備查詢，意即無法有效描述使用者本身真正的需求，導致搜尋引擎的回傳結果摻雜使用者不感興趣的非相關資料，使用者必須額外浪費時間做過濾的動作，大大降低搜尋的效能。有鑑於此，本研究實作一套搜尋引擎，將使用者目前瀏覽網頁的行為與時間匯整成使用者資訊，搭配相關回饋機制將使用者資訊回饋給系統，藉此分析出使用者真正的興趣，同時重新排序使用者尚未瀏覽的資料，優先輸出使用者真正感興趣的資料。
本搜尋引擎以線上即時訓練的方式，直接適應每個使用者對於不完備查詢的真正興趣，解決了一般個人化技術必須預先記錄使用者喜好的缺點，並以不改變使用者瀏覽習慣為原則，搭配相關回饋機制與重新排序之功能，成功提升了搜尋的品質與效率。實驗中分析分群演算法Vector Space Model與分類演算法Support Vector Machine套用在本系統之效能，發現採用VSM演算法能夠有效避免因人為因素而定義出錯誤的訓練資料集，且效能較SVM為佳，因此本適應性搜尋引擎採用VSM來處理資料之預測。

Search engines have become the most powerful assistants that search for useful data to people in this modern society. People often get what they need by input some queries to search engines, but most research results show that there is merely one input query submitted to the search engines by 77% users approximately, and most queries submitted by 85% users contain less then three words. According the above mention, most users provide “incomplete queries” to search engines, and these incomplete queries are not able to describe what users need; so that the results returned by search engines are peppered with many impertinent data, and users must pay more effort to filter the resulting data returned by search engines. Therefore, the efficiency of search engines is diminished enormously. In view of this situation, we proposed a novel search engine to cope with this problem. Our proposed search engine makes use of a user’s browsing behavior and interval to profile his information, and then combine this information with the mechanism of relevance feedback to analyze the user’s real interest; finally the search engine rearranges the resulting data which are not browsed by the user yet, so that the user’s real interest data are outputted in the prior positions. Furthermore, our search engine uses on-line training technique to adapt itself to each user respectively, and it profile the user’s information implicitly. After experiments, our proposed search engine promotes search quality and efficiency successfully.

[1]Silverstein, Craig, Monika Henzinger, Hannes Marais,and Michael Moricz(1998), “Analysis of a very large Log”,Digital SRC Technical Note 1998-014.
[2]Major Bernard, Amanda Sping, Tefko Saracevic,”Real Life Information Retrieval： A study of User Query On The Web”
[3] Sullivan, C., Web developer.com guide to search engines, John Wiley & Sons, 1998, pp.17-52
[4] Mauldin, M. I. Lycos ： design choices in an Internet search searvice, IEEE Expert, Vol. 12 1, Jan. – Feb. 1997, pp.8-11
[5] Kingoff, Comparing Internet search engines, Computer, Vol. 30 4, April 1997, pp.117-118
[6] Thomas, B., URL diving, IEEE Internet Computing, Volume： 2 3, May – June 1998, pp.92-93
[7] University of California, Berkeley, Meta-search engines, http：//www.lib.berkeley.edu/TeachingLib/Guides/Internet/MetaSearch.html , April, 2000
[8] Ellsworth, J. H., Working the net： time to consider a web search specialist, netWorker 3 3, Sep. 1999, PP.17-19
[9] Frankes,W. B. and Ricardo, B. Y., Information retrieval： data structure & algorithm, Prentice-Hall,Inc.,1992
[10]Salton, G. and McGill,M.J., Information to madern information retrieval,McGraw-Hill,New York, 1983.
[11] Jones,K. S. and Willett, P.,Reading in information retrieval,Morgan Kaufmann Publishers,Inc.,San Francisco,California,1997,pp.167-190.
[12] V. N. Vapnik. The Nature of Statistical Learning Theory. Springer, New York, 1995.
[13] Steve G.1998, Support Vector Machine for classification and regression. ISIS technical report, Image speech and Intelligent systems group of University of Southampton.
[14] O. L. Mangasarian and R. R. Meyer. Nonlinear perturbation of linear programs. SIAM journal on Control and Optimization, 17(6)：745-752, November 1979.
[15] D. P. Bertsekas. Nonlinear Programming. Athena Scientific, Belmont, MA, 1995.
[16] O. L. Mangasarian. Nonlinear Programming. SIAM, Philadelphia, PA, 1994.