AN IMPROVED PERFORMANCE MEASURE FOR USER PREFERENCE BASED SEARCH

Abstract

Today internet usages are rapidly increasing. Users are more dependent on the web search for needs of the information. In existing system the search engine provides the information that are needed for the user but it does not give the information that are relevant to the user’s interest. Search engine does not have much know about users interest. Personalization methods are introduced to improve the information retrieval system. In the proposed system personalization is done by three phases: a) Building user profile b) Applying re-ranking technique c) Performance measure. User Profile is an input for performing personalized web search. The user profile which captures the user interest are built from past search history of the user. User Profile contains basic profile and updated profile. We generate Profile score (PS) from basic profile score (BPS) and converged profile score (CPS). Basic profile score is obtained explicit from the user preferences whereas token extraction and count of hit links are used to calculate the converged profile score (CPS). By identifying the needs of individual user’s preference we can greatly improve the search engine performance. We fetch the top N results from the search engine and the weight score (WS) is calculated using term frequency and inverse term frequency (TF-IDF) The cosine score (CS) is calculated using cosine similarity. The cosine similarity is used to measure the similarity between the set of documents. The Re-ranking Score is calculated for providing the better results to the user. It can be calculated as sum of profile score, weight score and cosine score. The relevancy of search results is measured using tanimoto coefficient and levenshtein distance. Thus the qualities of search results are progressed.