Research On Key Problems Of Collaborative Filtering Algorithm In Recommendation System

With the development of Web technology in the Internet, the user is no longer simply to obtain information from the network, but to take a more active way to produce information. Due to the rapid growth of the number of users, user-centered information generation mode result in a rapid increase in the amount of Internet information, this phenomenon is known as "information overload". This phenomenon refers to that in front of the massive information, people can not quickly and accurately access to their useful information. In order to solve the problem of "information overload", the recommender system is produced. Recommendation system does not require users to provide accurate demand, but according to the analysis of the user’s past behavior to infer the user information that may be required in the future.At present, collaborative filtering recommendation technology has been widely used in electronic commerce because of its unique advantages. Although collaborative filtering recommendation algorithm has achieved many results, but there are still many problems need to be solved. Such as "cold start", "scalability", "data sparsity" and other issues, as the existence of these problems, the accuracy of the algorithm has been affected. How to solve the above problems and improve the performance of collaborative filtering algorithm has been the focus of the recommendation system.The main work of this paper is as follows:Firstly, according to the problem of "cold start" and "scalability" in collaborative filtering, a collaborative filtering recommendation algorithm based on clustering of user attributes (ID-CF) is proposed. The recommendation system combine item-based collaborative filtering with K-means algorithm by adding the weight of the method, significantly improve the accuracy of its recommendation, In this algorithm, the similarity between the project and the user clustering can be calculated off-line, which can solve the problem of scalability of the recommendation system. When a new user joins the system, through the use of clustering algorithm, to add the new user to the most similar to the user set, so we can quickly predict the user’s score on the project, the cold start problem can be solved.Secondly, due to "data sparsity "problem has a great influence on accuracy of collaborative filtering algorithm, a collaborative filtering recommendation algorithm based on the graph model (NG-CF) is proposed, the algorithm presents a new similarity measure standard, namely, similarity between users or items can be obtained through the relationship between vertices in the graph, and then use the k-nearest neighbor algorithm to produce a prediction. Experiments show that even if the data is sparse, the results also have good stability."Cold start" and "scalability, "data sparsity" and other issues are reserach hot issues of collaborative filtering recommendation algorithm research, the paper is on the basis of previous work, just to make some exploration and analysis, there are still many deficiencies need to be improved.