Learning Question Structure based on Website Link Structure to Improve Natural Language Search

• Main Authors: Kao-Hung Lin, 林高弘
• Other Authors: Wen-Hsiang Lu
• Format: Others
• Language: zh-TW
• Published: 2006
• Online Access: http://ndltd.ncl.edu.tw/handle/81246785633692981415

碩士 === 國立成功大學 === 資訊工程學系碩博士班 === 94 === The indexing method of conventional search engine always regards one page (or document) as primary indexing unit. Therefore, we cannot record the structural information between web pages. Furthermore, conventional search engines cannot deal with natural language questions effectively. For alleviating these two problems, first, we propose a structural indexing method to record the link structure information between pages. Second, we propose a Tri-link Mapping Model (TMM) to extract question implicitly structures embedded in user’s natural language questions. On the basis of question structures, we can map relevant Tri-links in the relevant website. We also propose a heuristic method to extract question structure by dividing question into three parts, and calculate the similarity with Tri-link based on the Position Dependent Similarity (PDS) and Position Independent Similarity (PIS). These two kinds of similarity measures are based on conventional Vector Space Model. The performance of structural indexing method and conventional indexing method will be analyzed according to our experiments. Experimental results show that structural indexing method can get higher precision than conventional indexing method for natural language search.