An Implementation And Optimization Of The Collaborative Filtering Recommendation Algorithm Based On Category Preference

The rapid development of Internet has brought explosive growth of information. Facing vast of information, the normal requirements of people have been covered. How to find the real needs of the current users among these information and how to find the real needed goods from all categories of goods have become an urgently problem for the current network users. Personalized recommendation system, as the most effective information filtering method has been applied to every field. Recommendation algorithm, which is the core of recommendation systems, has more and more attracted people’s attention. Collaborative filtering recommendation algorithm(CF), the most popular recommendation technology, is one of the hottest research topics for now. However, data sparseness, cold start and low accuracy of recommendation still restrict the performance of the collaborative filtering recommendation algorithm seriously.In this paper, in order to solve the data sparseness and to improve the accuracy of the recommendation algorithm, some researches have been done in the following aspects: Firstly, by using the Term Frequency-Inverse Document Frequency(TF-IDF) principle, the user-rating matrix is transformed into user-category rating matrix which reduces the data sparseness greatly. Based on that, the category collaborative filtering recommendation algorithm(GCF) is proposed. Secondly, combine GCF and Userbased collaborative filtering(User-CF) with dynamic weights, the hybrid recommendation algorithm(CGCF) based on category of item is put forward to obtain better recommendation performance. Thirdly, two kind of algorithms combining UserCF, GCF and Item-based collaborative filtering(Item-CF) together are proposed to improve the forecast accuracy. Those two combination algorithms give full play to the advantages and avoid the disadvantages effectively of each single model. They balance the global and local advantages of each single algorithm in combination algorithms.Finally, three experiments of CGCF, PSO-CF and BP-CF are conducted on Movie Lens dataset to evaluate the proposed models. The experimental results show that CGCF can effectively reduce data sparseness. It not only ensures the prediction accuracy, but also improves coverage largely. These advantages provide an promising research direction in mining potential long tail problems of users and items. Based on GCF, PSO-CF and BP-CF combination recommendation models have some further optimization processes. All of the three models improve the prediction accuracy of recommendation algorithm largely.