Algorithm And Framework Of Recommender System Combined With NMF

With the development of Internet technology, the amount of information on the network increase exponentially. Due to the mass of information, the phenomenon of " information overload " become emergence. Under this circumstance personalized recommender Systems were proposed. Recommender systems aim to provide people with information they may be interested in by analyzing the user’s personal information or historical access records. Personalized Recommender System has become one of the most effective tools to solve the problem of information overload, and has been successfully used in many websites.This thesis gives a brief review of the research of recommender systems through the aspects of applications, advantages and disadvantages, the main algorithm framework. And the main works of this thesis are as follows:Firstly, we introduce another matrix decomposition method-Non-negative matrix factorization. And combine this method with the traditional Item-based collaborative filtering recommendation algorithm and User-based collaborative filtering recommendation algorithm. The fusion algorithms can reduce the noise in the user ratings matrix, improve the accuracy and reduce time consumption. In order to further improve the coverage of the algorithm, we mixing the two kinds of methods together. It has been proved that the algorithm smaller loss in accuracy, but can greatly improve coverage. It is useful to the excavation of the "long tail".Secondly, we provide a generic recommendation system framework and then introduce its main modules from three levels, the architecture from the outside, the recommender system architecture, the recommendation engine architecture. The advantage of the framework structure is its multi-engines. Because of its independent of the engine structure, the system can be flexibly adjusted according to the algorithm. We introduce the framework of the recommendation engine for collaborative filtering algorithm combined with Non-negative matrix factorization.