Study Of To Improved Collaborative Filtering Algorithm Using Pareto Dominance

Vigorous development of the Internet makes people live in the era of data explosion, this era is full of huge information which as vast an ocean. People are faced with many choices every day. It is usually that they cannot find the satisfying information after spending a lot of money and energy. The appearance of recommendation technology helps people to find the information that interest them and understand their potential demands. Recommendation service can direct users to buy goods in online store, can recommend friends in social network, can recommend movies, delicious foods, books, etc., it bring great convenience to people's lives. Among all kinds of recommendation algorithms, for the established model of collaborative filtering method is simple, using users` rating data to analysis users` preference, not requiring rely on the attributes data of users or items, collecting data convenient, the recommended quality is better, collaborative filtering recommendation becoming mainstream of academic world and industrial community.On the basis of studying various collaborative filtering algorithms and combining to the existing experimental conditions, this paper selected user-based collaborative filtering method as the research object. After analyzed the existing problems of collaborative filtering, this paper proposed a method to solve the data sparseness. In this paper, it used Pareto dominance to perform a pre-filtering process eliminating low similarity with the active user. It improved similarity measure by considering the influence of proportion of common ratings, combining with the PIP similarity measure which composed of proximity, impact, and popularity.Experiments conducted on two different sizes of classical Movie Lens data sets, it used proposed method in this paper to compare with other improved existing frequently-used similarity measures,JPIP and PIP algorithms. Using MAE(Mean Absolute Error), precision and recall as the measurement. The result of experiments showed that proposed method which using Pareto dominance to perform a pre-filtering process and combining with the JPIP algorithm improved quality of recommendation comparing with other methods.Contributions of this paper are:(1)Using the Pareto dominant theory to pre-filter the users with low similarity, improved the possibility of finding the users with high similarity.(2)Comparing with the traditional similarity measures only using the shallow content showed by rating score, proposed method analyzed the deep meaning hidden in rating score to improve the similarity measures.(3)Proposed method achieved better quality of recommendation, alleviated data sparseness.