Research On Matrix Decomposition Algorithm For Personalized Recommendation

The rapid development of the Internet has provided great convenience for users' work,study,and daily life.Users can access more fresh information through the Internet at any time.However,due to the nature of the Internet,information overload has become more and more serious.At this time,the recommendation system came into being.Among them,the collaborative filtering recommendation algorithm uses its personalized recommendation method to mine each user's deeper interests and hobbies,and quickly emerges from many recommendation systems.The matrix factorization model based on the latent factor model has been extensively expanded and applied because of its accuracy,efficiency,and ease of implementation.The main research work of this paper are as follows:1.This paper compares and analyzes several popular personalized recommendation systems first,and summarizes the problems faced by today's recommendation systems by understanding the recommendation system.Then it makes a detailed summary and deduction of the evaluation and optimization of the recommendation algorithm.2.The first core section provides an in-depth analysis of matrix decomposition and foundation models from a structural model perspective.In order to further improve the prediction accuracy of the model,the scoring time auxiliary information is introduced on the basis of the decomposition factor matrix decomposition model,and defined as the decomposition of the time bias factor matrix,and the iterative formulas of the algorithm,and the algorithm structure diagram and pseudo code are deduced in detail.Then on the real data set,the influence of various parameters on the algorithm is determined by experiments and the optimal parameters are selected.Finally,the same parameters are fixed and compared with other algorithms to verify the accuracy of the improved algorithm.3.The second core section deduces decomposition models and extended of probability matrix from the perspective of statistics Based on this premise,on the basis of the probabilistic matrix decomposition model,it is introduced whether the item has a score attribute as a limit item,and is defined as the attribute matrix factorization of the user attribute.And detailed derivation of the algorithm's iterative formulas,as well as the structure diagram.Then the appropriate parameters are fixed according to the characteristics of the improved algorithm to further verify the pros and cons of the algorithm.Finally,the same parameters are set to iteratively compare various algorithms to verify the accuracy of the improved algorithm.