Research On Matrix Sparsity In Recommendation System

Compared with the serious shortage of information in the past,the current rapid development of the Internet makes the amount of data drastically expanded.However,in the face of massive data,whether the information can be accurately filtered and filtered is becoming an important indicator for measuring the pros and cons of an information system.As an information system for providing favorable experience for users,selecting and filtering the massive data and showing the most concern information for users is the core function of the recommendation system.Therefore,to solve the problem of filtering the information is one of the most important issues in recommendation system.Recommendation system can solve the problem of information’s overload,However,with the continuous expansion of the recommendation system’s scale,The user-item scoring matrix shows extreme sparsity,resulting in a serious decline in the recommended quality of the collaborative filtering recommendation algorithm based on the traditional similarity measure method.In this paper,the sparseness of the scoring matrix in the recommendation system is studied.The users’ characteristic information is introduced.Combined with the data’s preprocessing,similarity’s enhancement and mixed prediction,the sparseness of the scoring matrix in the recommendation system is solved.The research work and innovation of this paper include:(1)On the MovieLens dataset,we analyzed the user ratings and the common rating scale,In addition,we analyzed the sparse features of the scoring data from both theoretical and numerical aspects,and discussed the impact in the recommendation system brought by the sparse problem.(2)An improved similarity enhancement algorithm is proposed,which uses SVD to preprocess the initial scoring matrix to improve the initial similarity between users and items.Through the statistical analysis of the common scoring scale of the users and the items,take the product of the common scoring scale and the reference factor as the weight of the initial similarity in the similarity enhancement model.In the phase of predicting scoring,the user-based predictive scores and the item-based predictive scores are mixed with common scoring weight.The improved model is tested on the dataset.The results show that when the value of λ is 1.6,the error reaches the minimum value,and the minimum average absolute error MAE is 0.7302.It shows that the improved similarity enhancement proposal proposed in this paper can improve the quality of recommendation.(3)A similarity model of users’ features is proposed.By extracting and quantifying the users’ information in the dataset,a similarity algorithm based on ratio is adopted to refine the model of users’ feature and obtain the users’ feature similarity matrix,which is weighted to the initial similarity of users.The numerical analysis shows that the proposed algorithm can reduce the recommended error by about 0.01 after the users’feature is added,which can further improve the recommendation accuracy and alleviate the sparse problem.(4)The MATLAB simulation platform is adopted in the phase of experiment.The users’ rating information is imported to generate the user-item rating matrix,and the users’ feature is imported and quantified to refine the model of users’ feature.First we use SVD to do data preprocessing,and calculate the initial similarity,then start algorithmic iteration and blend the predictive ungraded items.The results show that the improved similarity enhancement algorithm proposed in this paper can further solve the problem of the sparse problem in the recommendation system.Combing with the model of the users’ characteristics,the recommendation’s quality can be better improved.