Research On Collaborative Filtering Recommendation Algorithm Based On Multi- Attribute Rating

With the continuous development of the Internet and e-commerce technology, the information resources on the Internet grow exponentially and vast amounts of information is presented in front of people. It is difficult to quickly find effective resources they needed, which is called the "information overload" phenomenon. In order to handle this situation, personalized recommendation system come into being. these personalized recommendation system will not only help users find the resources they needed, improve the utilization rate of resources, but also bring a better user experience and business opportunity for service providers. In today’s era of big data, personalized recommendation technology is widely regarded as an essential service in the field of e-commerce technology.In the field of personalized recommendation system, collaborative filtering algorithm is the most widely used and most successful technology. Its main idea is that according to the user’s historical similarity rating recommend to target users predict project which were based on the preferences of the project nearest neighbor. Although collaborative filtering recommendation algorithm has been widely used, we can found some defects, such as data sparse, cold start, low recommendation accuracy, lack of trust and other issues.In this paper, through in-depth studying in collaborative filtering algorithm, we found that collaborative filtering recommendation algorithm’s accuracy was not high and the user rating lacked trust. So this paper proposed the collaborative filtering recommendation algorithm based on multi-attribute rating. First, due to the traditional collaborative filtering algorithms were single overall rating, lacked the multifaceted understanding of user preferences interest, it resulted in that recommendation was not accurate enough. This paper introduced the project multi-attribute rating, and used the information theory entropy change of the user attribute rating amplitude to analysis. According to the change of amplitude of attribute rating, we can determine the proportion of the user’s own history rating and the rating of users who were target users’ neighbor. And then The comprehensive evaluation which was based on the two projects of composite rating determined the rating of each attribute. Finally, the users overall rating of the project would be computed by the rating of each attribute. By this method, it could solved the shortcomings in a single rating. For the trust of a malicious user attack problem, in this paper we made use of the deviation of user ratings and the number of items which were rated together to consideration of trust between users. We could filter out the lower trust neighbor from target user’s neighbors, which would isolate the malicious users to recommend a neighbor set, so that these methods can ensure the recommendation accurate and reliable.Finally, this paper regarded the current wider evaluation methods of recommendation system as the standard evaluation methods to experiment. The experimental results showed the proposed algorithm is practicable, and it whether on the recommendation accuracy or reliability was better than the traditional collaborative filtering algorithms.