Collaborative Filtering Recommendation System Research And Realization Of The Key Issues

With the rapid development of Internet and the growing wealth of information, information overload are increasingly serious. In this context, recommendation system comes into being and collaborative filtering technology is the most successful technology. However, with the expansion of application scope and the change of application environment, collaborative filtering technology gradually exposed some shortcomings, such as data sparsity, cold start problems, recommend low accuracy and so on. So this paper proposes two collaborative filtering algorithms to alleviate these problems.1. Collaborative filtering based on combination of expert opinionsWith introducing expert opinions and user attributes, the algorithm alleviates data sparsity, and ultimately achieves improving the recommendation accuracy. Firstly, we propose a novel similarity measure with user attributes in order to highlight users'background. Then, combining users'ratings with expert opinions, the prediction gets a right balance in both expert professional opinions and similar users. Finally, smooth the prediction by SVD-based collaborative filtering algorithm. Theoretical analysis and experimental results show that the algorithm improves effectively the recommendation accuracy.2. Collaborative Filtering Based on Time DivisionWith introducing degree of interest, the algorithm alleviates data sparsity, and ultimately achieves improving the recommendation accuracy and alleviating new item problems. Firstly, according to the released time of items and the current time, items are divided into three types, named new item, normal item and old item respectively. Then, the degree of interest in three types can be got with the time of item rated, by which user-item matrix is expanded. When predicting the rate of an item for a user, different strategies are used with different types of items. Theoretical analysis and experimental results show that the algorithm improves effectively the recommendation accuracy and relieves new item problems.Finally, on the basis of the above proposed algorithm, a model of movie recommendation system is designed and realized. We do fully analysis and conclude all parts of the system for laying the foundation of future theory and practice.