Collaborative Filtering Technology Research Oriented Sparse Data

With the rapid development of information technology and the popularization of mobile devices,information presents an explosive growth,and there is a problem of "information overload" and "information lost".How to quickly and accurately help users to pick out their interested items has become a difficult problem in the era of big data,and the recommendation system came into being.Collaborative filtering technology is one of the most widely used and successful recommendation algorithms in the recommendation system.It analyses the historical behavior of users and provides personalized recommendation service to users.With the increasing scale of commercial websites,users usually score only a small part of their projects,and the user rating matrix is very sparse,and the quality of recommendation is decreasing.In order to mitigate the impact of sparsity,the existing solutions are mainly null value fill technology and the use of new similarity computing methods.The idea of empty value filling technology is to use the existing score data to fill the missing value of the user.This method is easy to use and does not bring additional burden to the user.But the null value filling technology itself is a kind of artificial prediction for the missing value of the score,and the prediction value can not represent the real preference of users.The new similarity calculation method uses all kinds of information contained in the user's score data,such as the user's common scoring project,and the ratings of the users to calculate the similarity between users.This calculation method has high stability,but it is still based on the finite score data of the user and recommends the quality of the system.The degree of improvement is limited.In view of the above problems,the main work of this paper is as follows:(1)We propose a collaborative filtering algorithm based on Sparse Clustering and user trust.In view of the problem that the clustering effect is not ideal in the case of data sparsity and too much useful information is rounding when dimensionality reduction,the sparse subspace clustering algorithm is used to cluster the users,to retain more useful information,and to calculate the user trust in the similarity calculation: first,the validity of the user in the data set is calculated.The fairness score is used to set up the user's trusted degree matrix for each user,and then the improved user trust based on the scoring mode is fused,and the similarity calculation is carried out in the traditional similarity measure method.The experimental results on the movie dataset show that the algorithm can alleviate the problem of finding inaccuracy in sparse data and improve the recommendation quality.(2)A collaborative filtering algorithm based on user common interest score and score based time difference is proposed.In the field of movie recommendation,in view of the problem that the existing recommendation algorithms do not take into account the difference in scoring in the common scoring project,we set the data with little difference in the common score of the user to the common happy praise diversity.The existing algorithm of the recommendation system does not consider the effect of the user scoring time(view time)on the user's similarity when the movie shows,and proposes an algorithm based on the difference of the user's scoring time,and calculates the difference between the user and the user on the score time between the user and the user at the film screening stage,but then the traditional similarity calculation method is carried out.It is improved that the users' similarity degree is higher on the common score item.The experimental results on the movie data set show that the algorithm can improve the quality of recommendation.