The Research On Framework Model Of Recommender System And Collaborative Filtering Algorithm

In recent years, with the rapid development and maturation of the mobile devices, people are facing the times of information overload, where there are full of huge quantity of information, which make users very difficult to find what they are truly interested in. Recommender system aims at overcoming the problem of information overload, to free people from massive redundant information. It recommends possible information which users might be interested in, by analyzing historical information about the user to establish the interesting model. Because of this, recommender system has been widely used in the filed of application and research.The main contributions of this thesis are summarized as follows:1. The thesis focuses on recommender system framework model in terms of cold start. Existing cold start solution typically provides personalized hot tip recommendation for new users, yet in a coarse-grained way. With existing recommender system framework model and cooperation principles, the thesis proposes a cooperation-based recommender system framework. The cooperation among recommender systems drives to converge multi-source and heterogeneous data for users to solve cold start issues, improving the system accuracy and delivering proactive and transparent service.2. The thesis analyzes and emphasizes on collaborative filtering algorithm, especially the optimization of user-based collaborative filtering algorithm. It designs the process of extracting vectors on users preference to items, combined with similarity measures discovers a collaborative filtering algorithm based on user preference level(three preference levels, i.e. attribute level, behavioral level and score level), and makes simulation with dataset MoviveLens. The optimization method proposed in this thesis proves to be highly effective in enhancing the recommendation performance, whether in rating prediction or Top-N prediction.The survey can help alleviate legacy recommender system's cold start problem, providing models to support the system application. In addition, a highly accurate collaborative filtering algorithm could be achieved by user based optimization, technically promoting user satisfaction and advocacy on recommendation service. The survey lays a solid foundation for further application.