Study And Implementation Of Hybrid Recommendation System Based On Collaborative Filtering

With the rapid development of Internet technology, we have entered the era of information explosion. Faced with such a flood of information, we have to spend a lot of time to quickly find the information we need. Today many applications can help users to filter information such as portals, search engines, etc. But they don't consider individual needs, so it can not be a good solution to the problem of too much information. Personalized recommendation system can recommend information that users may be interested in according to each user's personal information. Therefore, it has become a way of solving the problem of information overload, which largely saves the users' time to filter information.Firstly, this thesis describes the research background and current status of personalized recommendation system, also it has a thorough research and analysis on personalized recommendation system algorithm which is mainly used recently. Recommender system consists of collaborative filtering, content filtering, knowledge-based filtering and hybridbased filtering. Currently, these recommendation systems are faced with many issues, such as data sparsity, cold start, scalability, and so on. These issues largely affect the performance of the recommender system. The data sparsity and scalability issues are two important factors which affect the performance of recommendation. This thesis is aimed to reduce the impact of these two factors on the recommended performance. Firstly, we improve method of nearest neighbor search, which is based on the memory collaborative filtering. The method search the possible of nearest neighbor by intersection operation and compute the similarity of users by similarity function. And it uses some users of higher similarity as the real neighbor of objective user and collects these users ratings to form a new matrix.Comparing with traditional nearest neighbor method, the dimension of the original scoring matrix is reduced, so does sparsity. Secondly we propose a new hybrid collaborative filtering recommendation algorithm, which is based on the improved algorithm. Then we use matrix factorization technique on the nearest neighbor matrix. Because the matrix factorization technique can solve the scalability problem, so the hybrid algorithm we propose solve the problem of data sparsity and scalability at the same time.Secondly, we plan a detailed experimental program to verify the proposed algorithm, the test data sets we uses is the Movielens data sets which are widely used in the recommender system field. In order to avoid the influence of parameters on the final result, we analyze the factors which affecting the recommender algorithm performance from all aspects, and then we do a lot of experiments to determine the parameters of the final experiment. By comparison with the traditional collaborative filtering algorithm after the experiment finished, we conclude: collaborative filtering-based hybrid recommendation algorithm has been greatly improved in terms of accuracy, thus verifying the validity of the algorithm proposed.Finally, we also design and implement a simple movie recommendation system, which is based on Java EE system structure, using Java programming language and My SQL database, the development environment is My Eclipse2014. We use hybrid recommendation algorithm proposed for the core part. Then we provide recommendations for each user when they access to the system and display the recommendations in the form of a list. Finally, we make a detailed summary of the hybrid personalized recommendation algorithms and systems, also we make a prospect on the work to be undertaken in the future.