Gaussian Distributed Probabilistic Feature Ranking for Personalized Search on Web
R.C. Suganthe · 2015
2 Abstract: Personalized search on web is becoming a reality due to popular mobile devices and widely available internet services. However, characteristics of the search on web, such as ranking, unpredictable user behaviors and issues related to automatic updating of user behavior profiles, present challenges in selecting optimal services for personalized search on web. Traditional personalization methods efficiently predict gender orientation search performance in an online manner. But, it may not always result in the best composite service because constant changes without ranking user behavior patterns and lacking automatic update of user behavior profiles makes the performance of service invocation an ineffective model. With the objective of developing an effective ranking for user result on web mining system, a framework Probabilistic User result Feature Ranking based on Gaussian distribution (PUFR-G) is proposed. The framework PUFR-G initially identifies the result features and analyzes them with user automatically updated profiles. PUFR-G simultaneously takes into account current feature frequency using the dynamic diversity re-ranking procedure. The frequency of the updated user profile is changed and ranking is carried out for obtaining search engine personalization result. PUFR-G attains the query result for different personalization using the Gaussian distribution function. The function produces mean vector and covariance matrix to identify rank rate for user query result with the height of the curve peak representing the highest rank value whereas the lower point denotes lower priority range. PUFR-G based on the Gaussian distribution function reduces the latency time on end to end process. Experiments conducted using Freebase Data Dump demonstrates that our framework can obtain better solutions than current standard personalization methods on web. Experiment is conducted on the factors such as precision ratio, execution time for feature ranking, personalized information search retrieval rate and ranking efficiency.