A New Bisecting K-means algorithm For Inferring User Search Goals Engine

Different users may want to search different goals when they submit some ambiguous query, to a search engine. The inference of user search goals can be very useful in improving performance of search engine. To conclude user search goals by analyzing search engine query logs a novel approach is proposed. First thing is that, we propose a framework to find out different user search goals for a query by clustering the proposed feedback sessions. Feedback session is built from user click-through data and can efficiently reflect the information needs of users. Second thing is, we propose a novel approach to generate pseudo-documents by using feedback sessions for clustering. For clustering a new algorithm which is bisecting K-means algorithm is used. At the end, a new criterion Classified Average Precision (CAP) is proposed to evaluate the performance of search enging. This criteria gives us value for k-means and bisecting k-means algorithm which shows that bisecting algorithm has better performance than k-means.