Modified Collaborative Filtering Recommendation Algorithm Based On Rating Prediction And Its Application In NERMS

With the development of Internet and E-Business, the Internet has been a service center of globality information, and also a major way of people gaining information. But the Internet information are increasing so rapidly that we should spend many times to search or browse the information which we need. Because of the distributing, heterogeneity and dynamic of Internet information, it is a challenge to develop a rapid, available and correct way of information search. Thus, Recommender System becomes a important researching filed of E-business.Personalized Recommender technology is a important researching filed of Data Mining, it provides different recommending services for different users. The Personalized Recommender System implements the function of making recommendations for users initiatively by collecting and analyzing the information of users and learning the interest and action of users. Personalized Recommender technology can attract more visitors by improving the quality of service and efficiency of accessing on a web sit.Collaborative Filtering Recommendation technology is very successful in the Personalized Recommender researching filed. It makes recommend- ations for a user by finding his some most similar neighbors using rating matrix, it is that making recommendations for a user by using other users'information. Collaborative Filtering Recommendation System searches some most similar neighbors of a user with Statistic technology, and then predicts the rate of this user on a item using the rates of his neighbors on this item, at last makes recommendations for this user.In this paper, introduced the research and implementation of Collaborative Filtering Recommendation Algorithm based on Rating Prediction, then we do some modifies on this algorithm, at last we apply the modified algorithm in NERMS to improve the service quality and accessing efficiency of NERMS.First of all, introduced some knowledge of Personalized Recommender System; it is composed of the definition of Personalized Recommender System, the steps of implementing a Personalized Recommender System and some Personalized Recommender technologies. Recommender System recommends some objects which users are interesting in based on the interest of users, the system is also called Personalized Recommender Systems. There are two major Recommender Systems in Personalized Recommender technology; they are Recommender System based on content and Recommender System based on collaborative filtering. There are six steps of implementing a Personalized Recommender System, they are user profiles, user information collecting and updating, information processing, resource profiles, personalized recommender and recommendation results display. The Personalized Recommender technology is used so widely that more and more people research it, and some new recommender technologies such as clustering analysis, Bayesian Net, Horting map, relation rules etc appear.Secondly, introduced the implementation of Collaborative Filtering Recommendation Algorithm based on Rating Prediction in detail. Because of the extreme sparseness of user'rating matrix, traditional measurement methods of similarity can't find users'neighbors effectively, and the quality of recommendation is also not good; there is a simple method that giving the resource which a user dose not rate a default rating, but not all of the users will give the resource which they do not rate the same rating, so this method also can not resolve the extreme sparseness problem of user'rating matrix. Collaborative Filtering Recommendation Algorithm based on Rating Prediction can resolve this problem well, because before finding users'neighbors, it gives the resource which users do not rate a predicting rating, and then can find users'neighbors correctly. There are four steps of the implementation of Collaborative Filtering Recommendation Algorithm based on Rating Prediction: firstly, initialization, constructing the users'rating matrix; secondly, rating prediction, giving the resource which users do not rate a predicting rating; thirdly, finding users'neighbors based on these predicting rating; at last, making recommendation for a user based on his neighbors.Thirdly, introduced two modifications on the Collaborative Filtering Recommendation Algorithm based on Rating Prediction. At rating prediction step, the algorithm needs to find the resources'neighbors, and it finds neighbors from all of the resources; when the number of the resources is very large, the performance of the algorithm will descend rapidly. The first modification of the algorithm is using clustering algorithm, it is that the algorithm parts the resources into several subclasses before finding resources'neighbors, and then finds a resource's neighbors in the subclass which the resource is belong to, so the range of finding a resources'neighbors can be reduced, and the result of finding neighbors is more correctly. The second modification is using users'interesting subclasses, it is that first the algorithm parts the users'rating matrix into several sub-matrixes based on users'interesting, and then uses the rating prediction algorithm on the sub-matrixes, so this modification also can reduce the rang of finding resources'neighbors and make recommendations more correctly.At last, introduced the application of Collaborative Filtering Recommendation Algorithm based on Rating Prediction in NERMS. This part is composed of information gaining, information processing, making recommendations and results display. In the information gaining, the recommender system gains the information from the we logs which contains the users'actions such as browsing, collecting and downloading, and then gives every resource a rating based on the users'action; In the information processing, the system combines the new information and the old information, and adjusts the resources'rating; In the making recommend- ations, the system calls the recommender algorithm to make recommend- ations for the users; In the results display, the system will return the recommendation results to users by showing them on the personalized page.The recommender algorithm proposed in this paper is developed in Eclipse, the Data Base is DB2, Web server is tomcat; By testing in NERMS so much times, the results of algorithm turn up trumps. There are also some lacks in the algorithm, such as complexity of the algorithm etc, but we believe that by the development of computer's software and hardware, the Recommender technology will be more and more perfect and the problem above also will be resolved perfectly.