A Movie Recommendation System Based On Hybrid Recommendation Algorithm

With the clipping development of technology of the Internet,various data in the Internet have shown exponential growth.While these data bring convenience to people,they also bring about the problem of "information overload".At present,personalized recommen-dation technology represented by collaborative filtering has become an important means to solve this problem,enabling people to efficiently obtain the information they need from mas-sive amounts of information.However,when the era of big data is coming,the development of technology of collaborative filtering recommendation also faces many challenges,such as the issue of data sparseness,the issue of cold start and the issue of recommendation accuracy and diversification.In order to solve the issues of data sparseness,recommendati on accuracy and diversification brought by the single collaborative filtering recommendation technology,a hybrid recom-mendation algorithm is came up with this thesis,which can effectively start from the user's interest characteristics to realize different recommendation appeals in terms of user diversity and personalization.The main works are as follows:1.The development history and research status of traditional recommendation algorithms and deep learning technologies in recommendation systems are analyzed in detail.The deficiencies in the current research are proposed,the reasons for these deficiencies and the solutions are given.This thesis introduces the relevant theoretical foundations in-volved in this thesis,then it introduces the basic ideas,theoretical background and practical applications of collaborative filtering algorithms,content-based recommen-dation and deep neural networks,respectively.2.A hybrid model was constructed by combining the LFM,DNN and factor decompo-sition machine.The model constructed the combination of low-order feature and the combination of high-order feature of users and movies for mining the complex implicit connections between them,then it realized the Top-N predictions for recommended movies.Therefore,when the users want to discover new movies,the recommendation list can be accurately filtered and sorted.3.The experiment uses the MovieLens data set and compares the way proposed in this thesis with unitary recommendation algorithms.The results manifest that the method proposed in this thesis precedes other methods in terms of MAE,indicating that this method has better prediction accuracy in the prediction stage and can successfully implement recommendations for new movies.