The Design And Implenmentation Of Movie Recommender System Based On Hybrid Collaborative Filtering

In the era of Web2.0 Internet is becoming more and more mature, the information grows explosively, which brings the problem of information overload. Personalized recommendation technology is powerful for solving the problem. Among all recommendation technologies, collaborative filtering is the most successful recommendation technology at present. However, the traditional collaborative filtering algorithm has the problems of sparse data and changes of user interest, which leads to the limitation of the performance of the algorithm.In order to solve the above problems, a hybrid collaborative filtering algorithm combined with attributes and time factors is proposed. In addition, taking into account that every day there will be a lot of new films released, the user is facing the problem of film information overload. In this case, the improved algorithm is applied to solve practical problems, and a movie recommendation system based on B/S structure is designed and implemented. The main research work includes the following aspects:Firstly, a new comprehensive item similarity measurement algorithm based on information entropy is proposed to dispose the data sparse problem. In this method, we fully use of the advantage of the project attribute, because this information is relatively stable and not affected by the sparsity of the rating data. When calculating the score similarity, the attribute similarity based on information entropy is also calculated.Secondly, in order to solve the problem of the change of user's interest, a model of user interest change is constructed, in which two kinds of weights is employed. Among them, the first time weight is inspired by Ebbinghaus?s memory curve, by exponential attenuation function to simulate the change process of user interest, and for different user groups to set different attenuation coefficients, thus, highlighting the importance of recent user behavior data. The introduction of second kinds of data weight aims to overcome the negative effects brought by the first weight: weaken early valuable behavior data.Thirdly, according to the design of the improved algorithm, we created a B/S based movie recommendation system, including database module, offline computing module, online intelligent recommendation module, system information management module, etc.. The system mainly uses Web Java development environment. What?s more, JSP and servlet related technology are used to implement. By this way of implementation, it not only provide a highly efficient and simple programming tools, meanwhile, it can also enhance the compatibility, independence and reusability of web pages.Finally, by the experiment, we analysis and compared the performance and effect of the differences with traditional recommendation algorithm, and discuss the impact of integration project attribute and time factor on recommendation algorithm.The experiment based on the dataset of movielens shows, compared with traditional recommendation algorithm, the modified algorithm can not only alleviate the problem of rating data sparse, but also can improve the accuracy of the algorithm. In addition, the final implementation of the film recommendation system in the effective management of the film information, furthermore, it can provide users with targeted movie information recommendation service.