Research On Collaborative Filtering Techniques For Information Recommendation Systems

With the fast development of Internet applications and the diversity of information, the amount of information in Internet is increasing explosively. To find the needed information from huge resources and provide voluntary personalized information services have become the common focus nowdays for the experts, scholars and Internet users. In this context, Personalized Recommender Systems have emerged.Collaborative filtering is the most widely and successfully applied technology in personalized recommender systems, which has become the focus of attention of the academic community. This dissertation takes collaborative filtering technology as the research topic and deals with the cold-start problem, the recommendation quality and scalability problems, the high dimension and sparsity problems, etc. The research works completed are as follows.(1) A review is given on the development of recommender systems. The concept of recommender system is introduced, and the classification of recommender systems is summarized. The basic system models of collaborative filtering recommendation systems are given and important steps are discussed, such as model representation and the assessment data collection, etc. Then, a classification of collaborative filtering technology is given, several typical collaborative filtering algorithms are investigated, and the research challenges are pointed out. At last, other related technologies are introduced.(2) A cold-start collaborative filtering recommender algorithm based on new user’s implicit information and Multi-Attribute Rating Matrix (MARM) is proposed. Through the new users’ implicit information collection implicit rating is completed, User Item Attribute Rating Matrix (UIARM) is used to measure the similarity between the users and to reduce the data sparseness effectively. The User Attribute Item Attribute Rating Matrix (UAIARM) is created. The new item attributes are matched with the user attributes in UAIARM, the user attribute with the highest score is regard as the necessary parameter of the recommended users, and thus a new item cold-start recommendation is achieved. Similarly, the new user attributes obtained by analyzing the browsing behavior of the users are matched with the item attributes in UAIARM, the item attribute with the highest score is regard as the necessary parameter of the recommended items, and thus a new user cold-start recommendation is achieved. And the extreme case for recommending a new item to a new user can be handled. In addition, the simplified UAIARM, MARM, can provide better cold-start recommendations with the relationship between the user attributes and the item attributes. Experiments show that the optimized algorithm proposed in this dissertatiom is effective for sparse rating matrix, especially for cold-start recommendations with new item and new user, the recommendation accuracy is mush better.(3) An improved KNN collaborative filtering algorithm based on clustering algorithm is proposed. When a traditional KNN predicts scores, the samples K nearest neighbors are treated equivalently, without considering the difference in the relevance between the K nearest neighbors and its category. Because the contribution of each of the different samples to the classification is different, it is necessary to treat the samples differently. The clustering algorithm is used to calculate the Classes-Relevancy, to differentiate K nearest neighbors of samples to be predicted. Experiments show that the optimized algorithm can improve the recommendation precision.(4) An improved collaborative filtering algorithm is proposed based on high dimension sparse rating matrix. Frist, the dimension of the high dimension sparse matrix is reduced. Then clustering is performed. Finally, the User-Fuzzy Cluster Rating Matrix (UFCRM) and the Classification Rating Matrix (CRM) are weighted to build K-Nearest-Neighborhood (KNN) set and to give prediction score and recommendation. PCA or SVD method is used for whole space dimension reduction pretreatment of the high dimension matrix to generate a low dimension matrix. An OPFCM algorithm is used for non-convex and arbitrary shape clusters. The improved algorithm optimizes the main parameters, avoiding the dependence of parameter setting. Experiments show that the OPFCM algorithm proposed in this dissertatiom has lower space complexity, lower time complexity and better cluster accuracy, and especially when the rating matrix is high dimension and very sparse, the prediction accuracy is much better.(5) A general purpose teaching resource recommender system is constructed based on CBR. The system framework model is designed, and a detailed explanation for the key steps of system is given, including the description of module design, development platform, and the development tools. The recommender system can prove the advantage of the algorithms proposed in this dissertation.