Research On Semi-supervised Learning Based Sparse Trust Recommendation

The recommendation system can provide accurate and novel recommendations for users,and effectively improve the benefits contributed by users.At present,it has been widely used in commodity recommendation,film and television recommendation,short video recommendation,news push,and many other industries.Collaborative filtering is a widely used recommendation algorithm in the current recommendation system.It uses users' historical browsing behavior to predict what current users may be interested in.However,collaborative filtering faces the problems of cold start and sparse evaluation data,which greatly reduces the performance of the recommendation system.To solve these problems,the trust relationship is introduced into the recommendation system.However,trust data often face the problem of data sparseness,which also affects its role in the recommendation system.This paper proposes the concept of sparse trust recommendation summarizes,and further improves the concepts of trust and sparse trust.In addition,the traditional trust binary representation method is not suitable for trust mining,and a new trust representation method is proposed in this paper.Specifically,this paper decomposes the two main sources(user preference and user behavior)that affect trust-building into four more fine-grained factors,namely similarity,consistency,reliability,and objectivity.Then this paper improves the Transductive Support Vector Machine algorithm(TSVM)to make it more suitable for sparse trust mining,and use it combine the trust influence factors to mine the implicit sparse trust relationships among users.After that,based on the Singular value decomposition model(SVD + +),we add the impact of social trust on users and the impact of sparse trust on users,and the influence of both is balanced to further improve the accuracy of recommendation model.Finally,the complexity of the model is analyzed,which shows that the sparse trust recommendation model has practical application value.The experimental design of this paper is divided into two parts,which are used to verify the effectiveness of the sparse trust recommendation model based on semi-supervised learning in sparse trust mining and user recommendation.Aiming at the sparse trust mining effect of the model,this paper selects three different trust mining models and compare them horizontally on the same four real data sets.In view of the user recommendation effect of the model,this paper selects six different types of recommendation models and compare them horizontally on the same four real data sets.The experimental results show that compared with other models,our model increases the density of trust data by 65% and the accuracy of recommendation by 3.3%.