Research Of Improved Collaborative Filtering Recommendation Algorithm

The rapid development of e-commerce has changed the traditional shopping mode. Compared with the traditional shopping mode, the online shopping mode is more convenient and efficient. However, it is also confronted with such problems as serious information overload of Internet, which requires users to locate goods of their own interest in massive amounts of commodity information. In this regard, it could greatly reduce users’ shopping efficiency and thus has become a major issue to be addressed in the current e-commerce.The personalized recommendation system features in customizing personalized recommendation service for users according to their personal characteristics in an effort to allow users to locate goods of their interest in a more efficient and convenient way. In this system, the personalized recommendation algorithm is the most important part. Currently, the personalized collaborative filtering recommendation algorithm is the most widely applied and researched recommendation algorithm, but the algorithm itself has such problems as sparse data and cold start which become the main concerns of the current research on collaborative filtering algorithm.The main work of this paper is as follows:(1) In the traditional collaborative filtering algorithms, the rating data prediction of target users relies heavily on the neighboring users’ rating data without considering the rating data characteristics of target users themselves. To this end, this paper proposes a prediction method based on the RBF (Radial Basis Function) neural network.(2) The similarity calculation method in the traditional collaborative filtering recommendation algorithm tends to cause the accidental error in case of sparse rating data. This paper adopts an effective similarity calculation method in which the percentage of common items numbers between the two users in the total number of all items that both users have conducted rating data as well as the original similarity calculation is taken into consideration in an attempt to improve accuracy of similarity calculation.(3) In the traditional collaborative filtering recommendation algorithms, the calculation is carried out based on either users or items without considering the combination of the two collaborative filtering recommendation algorithms. This paper adopts a hybrid user-item collaborative filtering recommendation algorithm which combines both the user-based and item-based collaborative filtering recommendation algorithms through a scaling factor to predict the rating data.(4) With the improved algorithms based on the RBF neural network and the hybrid user-item collaborative filtering algorithm, this paper presents a new collaborative filtering algorithm called the hybrid user-item collaborative filtering algorithm based on RBF neural network. Research findings show that the mean absolute error (MAE) value of the improved algorithm is smaller than that of the traditional algorithm and thus contributes to personalized recommendation for users in a more efficient way.