Research On Collaborative Filtering Algorithm Based On Improved User Similarity

The core of collaborative filtering is to predict and recommend unrated items to target users by historical rating information.The memory-based collaborative filtering algorithm is one of the most successfully representative algorithms,but there still have some problems such as high-dimensional sparsity and unreasonable similarity calculation.To solve the problems of sparseness and similarity in the traditional collaborative filtering algorithm,the thesis studies from the following aspects:(1)To alleviate sparsity in collaborative filtering algorithm,an improved method of data filling is proposed.By selectively fills user ratings in the process of similarity calculation,the method can effectively reduce the effect of data sparsity on the similarity calculation,and improve the recommendation quality;(2)To solve the phenomenon of asymmetric user influence in some social networks,this thesis proposes a new concept of asymmetric similarity matrix,which uses the influence factor and offset factor to adjust the traditional similarity matrix to form asymmetric similarity matrix.Meanwhile,the multidimensional group information is reduced to a one-dimensional virtual user with the same user characteristics.The approach provides a highly scalable solution to group and user hybrid recommendation.The experimental results show that the proposed algorithm can significantly improve the accuracy with having asymmetric users in the data set;...