Research On Social Network Active User Recommendation Based On Trust Calculation And Influence Factor

With the development of society and the progress of technology,a large number of social platforms emerge on the Internet,which will make a large amount of information available.In the social network with fast information transmission speed and large amount of data,the recommendation system can effectively predict the information needed by users and provide help to users.But compared with massive effective information in social network,the information received by users is little,so the sparse information is not conducive to mutual recommendation between users,which causes lower recommendation accuracy and worse user experience.Therefore,how to process the sparse information is important in current researches about social network.To solve this issue,the specific work of this article is as follows:(1)A collaborative filtering algorithm based on active user trust calculation and similar users are proposed for the sparseness of the trust matrix.Calculate from the user's activeness,professionalism,and credibility to measure whether a user is active.Then,use active users to fill the original sparse trust matrix and then a new trust matrix is obtained.Then use the improved Pearson coefficient to calculate similar users.Finally,the user's trust value and similarity value are used as weights,and calculate the sum of the scores of all trusted users and similar users,and a prediction model of the score is obtained.(2)A matrix decomposition recommendation algorithm based on impact factor is proposed for sparse user item scoring data.Firstly,improve the Pearson coefficient based on the trust value to calculate the similarity between users,then the improved Pearson similarity is fused into the original trust matrix,and the weighted trust matrix WTM is obtained.The conditional distribution of user's score and potential characteristics are constructed respectively,and then the influence factors are added.The influence factors is fused with the weighted trust matrix WTM,and finally the expanded probability matrix decomposition model is obtained.Finally,the two proposed algorithms are tested on the real sparse data sets Epinions,and then they are compared with other classical algorithms according to Mae and RMSE measurement methods.The experimental results show that the two proposed algorithms can effectively alleviate the problem of sparse data inthe recommendation system and improve the accuracy of recommendation.