Research Of Collaborative Filtering Algorithm Based On User Favorite Item Type

With the explosively growing of the technologies and services of the Internet, the information data world increases a lot and very rapidly. The expansion of information data has brought many issues such as Lost in Internet and Information Overload which become more and more serious every day. The methods to resolve Information Overload can mainly classify two kinds:Information Retrieval and Information Filtering. Based on the needs of the people, Information Retrieval will search out the related information and technology from the information stored in some way. Information Filtering will choose out user preferred personalized information from the dynamic information flow based on the filtering requests of the people.Recommendation technology is one of important application of Information Filtering, which has been applied broadly in E-economy area and drew more and more attention of researchers. Our research focuses on recommendation technology too.Although recommendation technology has developed several tens of years, but there are still challenges encountered, such as low precision, flexibility and real-time, high sparseness. The aspects of the research for recommendation technology can be divided to two main trends:filtering recommendation based on content and filtering recommendation based on collaborative process. Also, there are some other traditional recommendations technologies like recommendation based on association rule, clustering. We analyzed the advantages and disadvantages of these methods in detail. Then, we introduced the Collaborative Filtering recommendation algorithm (CF). Based on the user's preferred recordings, CF algorithm recommend products to this user. The current situation and clarification of the research of this algorithm has been exposited in our paper.Based on the analysis of CF algorithm, we proposed an algorithm of Collaborative Filtering based on User Favorite Type (UFTB). We figured out the main disadvantages of the traditional CF algorithms. The traditional Cosine-based Similarity and Correlation-based Similarity methods have not evaluations about the Null ratings in the user rating chart, which will do hurt to the recommendation. With regard to the bad sparseness of traditional collaborative algorithm, we proposed a new algorithm named as User Favorite Item Type based algorithm to address this issue. With this new algorithm, the space of items has been cut off efficiently and the operating efficiency has been improved. Regarding to the Null ratings, we used the item based collaborative filtering algorithm to handle it. And then, we try to prove the probability and validity of these algorithms. At the same time, we also proved our proposed UFTB algorithm.We did our experiments independently and synthetically. First, we compared the algorithm based on user favorite item type and the traditional collaborative filtering algorithm. Then, we compared the traditional collaborative filtering algorithm and the algorithm based on Null ratings predicted by the item based collaborate filtering algorithm. Finally, we bound these two methods proposed by us to UFTB algorithm.