Based Collaborative Filtering Recommendation System

With the rapid development of the internet, e-business is widely used.The system scale become larger and larger.Tradicitonal recommendation system can not satisfy the demand of giving the customers personalized recommendation about the products from so much information.Collaborative filtering is a popluar technique for reducing the information overload.However,most of the current collaborative filtering algorithms has the limitation of data sparsity.How to improve the performance of the prediction accuracy when the data is sparse is the major problem this dissertation study.In this dissertation, the data sparsity existing in the collaborative filtering system has been analyzed.we present the novel hybrid recommendation algorithms combining the singular value decomposition with user-based and item-based collaborative filter methods.The main task are as follows:1. Discussing the study significance and research status at home and aboard of recommendation system.2. Introducing the user-based and item-based collaborative filtering techniques in depth, then we discuss the advantage and limitation about this two techniques.3. Introducing the basic concepts of singular value decomposition and the evaluation metrics for the performance of personalized recommendation system.4. Studying the solution of problems about the data sparsity and similarity in depth.we first analyzed the current solutions, then, we present the HybridSVD method.The HybridSVD method first apply the single value decomposition method to get the neighborhood of the active users, then it uses the modificatory similarity method and the user-based and item-based collaborative filtering technique to obtain the final prediction of the active users.5. Comparing and analyzing the experiment result data which the HybridSVD and other traditional method on two publicly available datasets.The result show our metho have the better performance than others.