Research On Personalized Recommendation Based On Collaboration Filtering

With the rapid development of the Internet, the network data increase sharply,which leads the information overload and information confusion problems increasingly prominent. It's hard to find the accurate information for the users. Under this circumstances, the personalization recommendation system appeared. But the interest migration, data sparse, scalability and other issues impact recommended accuracy results seriously, how to solve these problems is the focus of current research.Attend to the interest migration, data sparse and scalability, this paper do some improvement for the collaborative filtering technology, the main work is:(1) The recommended results of collaborative filtering are always impacted by the users' interest migration, so this paper build an interest migration model based on the users' interest about item styles. The model considers the user's interest to the item style and uses the rate time to decay the scores of different item styles.Then calculate and predict the user's similarity based on the attenuated scoring matrix,alleviate the impact of interest migration on the accuracy of recommend result.(2) Build a trust model for collaborative filtering algorithm, which considers the time factor and the number of ratings for user's trust, use T-adoption and reliability metrics to calculate the degree of trust between users. Then combine the trust matrix with the time-based user similarity matrix. Recommended by the comprehensive weight can alleviate the impact of sparse data on the recommended.(3) As for the scalability problem, we use K-means clustering algorithm to cluster users.Then put forward a calculation method based on the entropy of the attribute similarity. The method improves the similarity of attribute accurately by considering the different contribution of different attributes.(4) In this paper we designed a series of experiments to verify the algorithm based on the Movie Lens dataset. Experimental results show that the proposed improved algorithm can improve the recommendation accuracy and scalability.