Personalized Recommendation System Based On Collaborative Filtering Technology Research

With the rapid development of the Internet Technology and Application, recommendation system has been widely used in the aspects of online shopping, book, music, movie, news, etc. For the vast amounts of information, the recommendation system is particularly important, and recommendation technology has gained more and more attention and research. Meanwhile, The recommendation algorithm is the core of the recommedation system, and in a large part, it determines advantages and disadvantages of the recommedation results. Collaborative filtering is the most popular and successful used one because of its simplicity and effectiveness.This thesis discusses the significance of recommendation system, research status at home and aboard, and some recommendation technologies such as association rules-based recommendation and content-based recommendation. This thesis deeply introduces the implementation process of collaborative filtering recommendation, its classification, and the main problems faced.This thesis proposed a hybrid collaborative filtering algorithm based on user and item against the sparsity and cold start problem of traditional collaborative filtering algorithms. To make the proposed algorithm be more representative of the user’s preferences and get the better accuracy prediction, this thesis mainly includes the following works:1. Introduce a new adjusted parameters a in order to modify the tranditional similarity measure formula;2. Introduce the control factor λ calculate weighted averages in the scores of user-based algorithm and item-based algorithm to get the better accurate prediction and ultimately improve the quality of recommendation system.This thesis use the MovieLens datasets which are the most used in recommendation system to design the corresponding experimental program. Firstly, design experiments determined the best value of the adjusted parameters a and the control factor λ. Secondly, do the recommendation effect test for the proposed similarity measure formula, and choose the different similarity formula for user-based and item-based algorithm. Finally, the experiment considers MAE as the evaluating indicator to compare the improved algorithm UPIC proposed in this thesis and commonly used algorithms. Experimental results show that the proposed hybrid collaborative filtering algorithm based on the user-item compared to the traditional collaborative filtering based on user and project algorithm recommended is more precise and gives better prediction in accuracy.