Research On Recommendation System Algorithm Based On Neighborhood And Matrix Factorization

The explosive development of the Internet brings a lot of information and data to users.However in the face of vast amounts of data,the user can't filter useful information from vast amounts of data,and utilization rate of information has actually dropped.Therefore,the filtering and efficient processing of information are becoming a general concern for users.The recommendation system came into being,and it is widely used to make intelligent recommendation of commodity for users in major e-commerce sites?Recommendation system helps users to save time and improve the efficiency of information utilization.Recommendation system as a branch of information filtering technology subject,it is widely researched and applied in recent years.However,the problems of the recommended system are going to be solved at present.This paper study the problems of the data sparseness in deeply,and put forward two improved algorithms.The main work of this paper are as follows:1.A collaborative filtering algorithm based on PC A and joint neighborhood weights is pr opose:Firstly,the algorithm reduces dimension by using PC A dimension reduction technique and transforms the original scoring matrix into the principal component space,effectively alleviating the data sparsity problem and preserving the dimension most representative of the user's interest matrix;Combining with the joint neighborhood weight model,which makes the similarity that based on the product more accurate and ensures the nearest neighbor's high-precision selectivity,reducing the time complexity of the operation.Experimental results show that the algorithm improves the recommended quality.2.The K-neighbor matrix factorization recommendation system algorithm is proposed.The algorithm combines the advantages of the classical matrix factorization and the K nearest neighbor algorithm.Considering the nearest neighbor of the target user and the target item When predicting the unknown score,the influence of the relevant score of unrelated users and unrelated items is greatly reduced.New definition of the neighbor score matrix and a new method of calculating unknown scores are proposed.The users and items neighborhood score information are merged into a neighborhood score matrix,and the rating information is obtained by mining the target user information.Experimental results show that the proposed algorithm is more accurate than the traditional collaborative filtering recommendation algorithm.