Research On Recommendation Algorithm Based On Matrix Factorization

In recent years,with the rapid development of the Internet,especially the mobile Internet,the prosperity of e-commerce has been greatly promoted.E-commerce platform,whether it is the number of goods or the number of users has shown explosive growth.How e-commerce providers can quickly and effectively communicate product information to target users,and how users can quickly and accurately find useful information from a vast amount of product information has become a hot topic of current research.In this case,the user's various behavior information can be used to accurately predict the user's point of interest,and the recommendation system that recommends the content of interest to the user is greatly valued.The accurate recommendation generated by the recommended algorithm helps the user to quickly find the commodity of interest,and converts the potential purchasing power of the user to the actual purchasing power as much as possible,which greatly increases the sales of the website.However,with the increasing scale of products and the number of users,the computing power of computers has brought great challenges.The matrix decomposition algorithm can decompose the high-dimensional user rating matrix into several low-dimensional matrixes,which greatly reduces the computational complexity and greatly improves the timeliness of the recommendation system.Moreover,the application of matrix decomposition can make the recommendation accuracy of the recommendation system continue to increase with the increase of scoring data,which increases the system's learnability.Therefore,the recommendation algorithm based on matrix decomposition has received more and more attention.In this paper,we have done some research on how to further optimize the recommendation effect of matrix decomposition recommendation algorithm.In order to solve the problem of the impact of neglecting neighboring users on the target users in the matrix decomposition recommendation algorithm,it is proposed that the neighbors of the target users should be found and given influence weights.At the same time,the matrix decomposition recommendation algorithm does not consider the problem of user interest drift,and gives the target user's interest item a time weight.In response to the above problems,the main research contents of this paper are:(1)For the problem that the target user is affected by the neighboring user,the influence of the neighboring user is added when the matrix decomposition is performed.In many user groups,each user's influence on the target users is different,similar users have greater influence,and users with greater differences have less influence.When the system uses the matrix decomposition algorithm to recommend,it should fully consider the influence between users.(2)How to find the target user's neighbors.This paper points out the disadvantages of the traditional KNN algorithm in finding the nearest user of the target user.It proposes to use Cloud-model instead of cosine similarity to find the nearest neighbor of the target user in a statistical sense.(3)The target user's interest drift problem is solved.The interest of target users changes over time.Some have long-term interests and some have short-term interests.It is obviously unreasonable to consider only the long-term interest of the user or only recommend the short-term interest of the user.Therefore,this paper integrates the time weights of target user interest into the matrix decomposition recommendation algorithm,integrates long-term interest and short-term interest to give project recommendations,and improves the accuracy of the recommendation system.