Research On Collaborative Filtering Recommendation Algorithm Based On Matrix Factorization

Due to the rapid development of the Internet and other information technologies,information overload has become a common problem in our big data era.Recommendation Systems play a vital role in solving the problem of information overload.However,the rapid increase in the amount of data accessible on the Internet has led to data sparseness and cold start problems,which means that clear rating feedback is not always available,and mainstream recommendation algorithms ignore recent preferences for user similarity and ranking of recommendation results Impact.These problems have become the bottleneck restricting the accuracy of collaborative filtering algorithm recommendation.Based on the above questions,this article conducts an in-depth study and considers more information related to users,such as non-joint rating items among users,the order of items when they are consumed and the entire interaction mode between users and items.Improve the user similarity calculation and matrix decomposition method of collaborative filtering recommendation algorithm.The main content of the thesis research is as follows:(1)The basic principles and classification of the recommendation system are explained in detail,and the characteristics of the model-based Collaborative Filtering recommendation algorithm are studied through the analysis of the traditional similarity algorithm and the in-depth interpretation of the matrix decomposition method.For the basis,the basic solution process of matrix factorization model is given.(2)In order to solve the problems that sparse user rating information density leads to low recommendation accuracy in the traditional collaborative filtering recommendation method,a probabilistic matrix factorization recommendation algorithm based on user trust is proposed.The algorithm first integrates the joint rating items and non-joint rating items between users,uses KL-divergence to obtain ranking of user trust which makes the feature vectors of similar users closer,and maintains this relationship during the probability matrix factorization process to improve the prediction accuracy of the model.(3)In order to solve the problems that sparse user rating information density and unsatisfactory ranking of recommendation results,a deep matrix factorization recommendation algorithm based on the longest common subsequence is proposed.The algorithm uses user information as a sequence,and uses the Longest Common Subsequence algorithm to find similar users,mske the recommendation results that are more in line with the user's interest appear in a higher position,and the hidden feature vectors among similar users are closer.And when using the deep neural network to learn the hidden features between users and items,it combines similar users to further improve the prediction accuracy of the model.Finally,a comparative experiment was carried out on the MovieLens 1M and Epinions datasets.The two collaborative filtering recommendation algorithms based on matrix factorization in this dissertation have excellent performance in various performance indicators,which fully verify that the algorithm proposed in this dissertation is in sparse data.The effectiveness of the recommended effect on the set.Figure[19]table[7]reference[75]...