Research On Movie Recommender System Based On Collaborative Filtering

With the rapid development of the Internet,the Internet has brought massive information to users,which satisfies the users' information requirements in the information age.However,due to the substantial increase in information,the problem of network information redundancy is serious.It makes users can not access to required information from a mass of data.How to get the required information and goods through the user data is worth studying,the recommendation system comes into being to solve this problem.Collaborative filtering recommendation technology is current one of the most widely used technologies in the recommendation system.The main idea is to find similar users or items for recommendation.The SVD and Slope One algorithms are collaborative filtering algorithms that can effectively make a recommendation prediction.However,there is a large amount of text data that have not been effectively utilized in the recommendation system,such as user comments,user tags.The disadvantages such as imbalance and sparsity in user data are also problems in the recommendation system.To solve these problems,the main work of this paper is as follows:Firstly,the principle of Slope One algorithm is explored.Aiming at the problem that the Slope One algorithm ignores the similarity relationship among users,we make experiments analyze the influence of accuracy of Slope One algorithm introducing the user relationships described by different user similarities.A kind of weighted user similarity calculation method is proposed and then the prediction accuracy of the Slope One algorithm by introducing the user relationship is impoved.Secondly,the principle of SVD algorithm is deeply studied.Aiming at the problem that user sample imbalance was not considered in the SVD model,the Slope-SVD algorithm is proposed to enhance the missing user data.At the begining of the Slope-SVD algorithm,the problem of user sample imbalance is handling.And then the sparsity of matrix is also weakened to some extent so that the accuracy of prediction and the availability of the algorithm are increasesd.Due to the simplicity and efficiency of the Slope-SVD algorithm,it is suitable for scenes with high real-time requirements.Thirdly,the traditional collaborative filtering algorithm ignores the text label data.To address this issue,the Word2 vec algorithm is used to integrate the user and movie text labels into the training process of the Slope-SVD algorithm proposed in this dissertation,providing more effective data for the training process.What's more,according to the case that the information amount provied by the hidden factor and the text label factor respectively in each sample is inconsistent,the weight vector is added for selection.The improved algorithm experiments are desingned in data set of MovieLens to illuminate a good prediction accuracy by comparing MAE value of different algorithms.