以Ontology為基礎動態學習語意查詢

Abstract

當搜尋結果傳回超過數百筆符合資料時，使用者常只瀏覽最前面幾十筆資料，雖然前幾十筆資料之後可能會有更有用資料，所以應該以搜尋結果前幾十筆資料的正確率來衡量搜尋品質。

本論文使用以ontology為基礎動態學習語意查詢提高搜尋品質，使用ontology記錄字與字間關連來儲存搜尋詞語意上差異，在搜尋核心Lucene查詢搜尋詞之前，先選出ontology結構中與搜尋詞相關連字詞來組成新的查詢，由觀察使用者點選行為動態決定新增查詢的權重。

When facing a lengthy list of search results, people often limit themselves to the top ten items on the list although there may be more useful information after the top ten items. As a result, the improvement of the search experience should be measured in terms of the precision rate of the top portion of the list.

We propose an ontology-based adaptive semantic search to significantly improve the search experience. To capture the semantic difference of search terms, naïve ontology is used to store the relationship among terms. Before a search term is processed by the search engine Lucene, the related words of the search term are selected from ontology structures to form new query phrases in the process of query expansion. The weighting of the expanded query phrases is dynamically learned by observing the users’ clicking behavior.