A NOVEL APPROACH TO INFER USER SEARCH GOALS FOR OPTIMIZE RESULT

Search engine is one of the most significant applications for internet users. Different users may have different search targets when they submit broad-topic to a search engine. Most times, search engine does not deliver what user needs. So to produce best relevant results, there is necessity to analyse user goals behind searching. Inference and analysis of it can be very useful in improving quality of a search engine's results. In this paper, a framework is projected to examine user search areas for a query by clustering the feedback sessions. Feedback sessions are made from user click-through logs and can more accurately predict the information requirements of users. Post feedback session formations, pseudo-documents are generated which better illustrate feedback sessions for clustering. For clustering, bisecting K-means algorithm is used. After cluster formations, restructuring of web search results is done by calculating smallest distance. At the end, to evaluate the performance of inferring user search goals, a new criterion “Classified Average Precision (CAP)” is proposed.