Research On User Preference-based Collaborative Filtering Algorithm

With the advent of Web 2.0,the amount of information grows rapidly on the Internet,and more plentiful data is available for users.Therefore,users have to consume a huge amount of time to obtain the valuable information.Especially the advent of the era of big data,the information overload problem has become the inevitable difficulty for scientific research.The recommender system is an effective resolution to solve the information overload problem in the large websites and e-commerce systems,which first analyzes the favorite of users according to the user record information,and then achieves the personalized recommendation.Collaborative filtering is one of successful technologies in recommender systems,but this technology is confronted with the serious problem of data sparsity,which restricts the improvement of recommendation performance.Hence,in this dissertation recommendation models based on user preference are constructed to improve the processing capacity for sparse data.The main contributions of this dissertation are as follow:(1)Some key scientific problems of collaborative filtering are summarized,and the causes for these problems is analyzed.In addition,the research on these problems at home and abroad is reviewed,especially,the research on the data sparsity is described in detail.(2)Aimed at taking user preference into account in the traditional collaborative filtering method.Therefore,a novel collaborative filtering algorithm based on user preference clustering is proposed,which first introduces clustering analysis based on different user preferences,and then defines a new similarity measure method to generate accurate and reasonable user groups.The experimental results show that our proposed algorithm can improve the recommendation performance.(3)An effective modified similarity measure method is proposed on the basis of the traditional collaborative filtering method,which modifies the traditional method from three aspects: user preference,the number of common rated items,and the similarity enhancement between users.The experimental results on two benchmark data sets verify the superiority of our proposed algorithm.(4)The traditional neighbor selection based on single similarity computation is not much accurate and reliable.Therefore,an information entropy-based collaborative filtering is proposed,which first introduces information entropy to describe different user preferences,and then adopts the large margin method to combine the information entropy and the similarity of user,moreover,selects high quality neighbors for active users.The experimental results show that our proposed algorithm can further improve the recommendation quality.