A Study Of User-based Collaborative Filtering Recommendation Algorithm

With the rapid development of the “Internet plus” and the advent of big data era,online shopping has become an important part of people's daily life.In the face of big data era of information display and push this urgent need to solve the problem,the recommendation system based on the user's personal preference,to provide them with a variety of different products or services recommended,so that the majority of users can quickly and accurately find their own satisfaction with the business or service.However,with the rapid growth of the number of Internet users and the rapid development of mobile Internet,although the recommendation system has made a great progress recently,but there are still some defects and deficiencies in the recommendation system.In this thesis,as user collaborative filtering recommendation system faced with the sparsity,scalability,cold start and other issues,through the singular value decomposition and clustering fusion,and improve the similarity measure formula and Top-N recommended weight value evaluation factor,A collaborative filtering recommendation algorithm based on singular value decomposition and clustering SCW(SVD Clustering Weight)algorithm is proposed.SCW algorithm is mainly based on the fusion of the singular value decomposition of dimension reduction,clustering and Top-N recommendation three principles.The concrete research steps can be summarized as follows: firstly,the thesis uses the singular value decomposition to construct a topic space from Movielens data set,and then calculate the similarity in this space.Secondly,the users of the dimension reduction score matrix are clustered,and the clustering analysis divides the data into meaningful clusters.Finally,after the completion of the clustering,according to the user's clustering of each cluster is similar to the behavior of the user,users of the target users to traverse the object set of other users,to remove the target user's article set to generate a recommendation list.And then,by using the improved predictive score,the prediction score of the recommended items is calculated,and the ranking is recommended.In this thesis,we use the Moivelens open data set to carry out the experimental verification,and compare the proposed algorithm with the traditional???correlation based collaborative filtering algorithm and the new algorithm for solving the scalability,sparsity and cold start up.The experimental results show that the proposed algorithm can effectively complete the personalized recommendation,in a certain extent,improve the accuracy of the recommendation and reduce the absolute value error,solve the problem of sparse,extended and cold start problems in collaborative filtering recommendation system.