Research On Collaborative Filtering Algorithm For User Rating Data

At present,the recommendation system has been widely used in shopping website,online music,movie entertainment website,etc.It can find every user's preference and give personalized recommendation.In the field of recommendation system research,neural network is used to obtain text,picture,voice and other features of the project,and use these features as auxiliary information for recommendation.Considering that user-item rating matrix is an important basis for collaborative filtering algorithms to make recommendations,this paper applies neural network to the analysis of user-item rating matrix,and proposes a method to obtain the characteristic vector of rating matrix and a hybrid recommendation algorithm.The main work of this paper includes:1)In this paper,we put forward using rating matrix row vector of each sample with the column vector matching way as the sample feature vector,and then use the BP neural network to rate prediction,so as to realize with deep learning to find the relation of potential user-item rating matrix.2)Considering that sparse user-item rating matrix will affect the accuracy of Slope One algorithm recommendation,a hybrid recommendation algorithm based on BP neural network and Slope One is proposed.The impact of rating matrix sparse on Slope One algorithm is alleviated by using BP neural network to predict project scores,so as to improve the accuracy of recommendation.3)With the algorithm proposed in chapter 3 as the core,a web-based movie recommendation system is designed.This system is based on the information of douban movies crawled.Users can browse the information related to movies on the website and give a score.When users have scored more than 5 points on the website,the background system of the website will recommend movies to users according to the collaborative filtering algorithm proposed in chapter 3 of this paper.On the basis of Movielens data set,a simulation experiment was carried out on the rating prediction method and hybrid recommendation algorithm proposed in this paper,and a comparison experiment was conducted with the open source tool Surprise on the value of RMSE and MAE.