Research On Anti-sparsification Trust Based On The Trust Source Of Textual Ratings

With the increase of online business interaction and the rapid development of the Internet,trust data has been widely used in recommendation systems.Trust data can help improve the prediction accuracy of recommendation systems and help users find relevant information.However,on the one hand,it is difficult for users to get the data they want because most of the data for specific users and tasks is garbage and noise.On the other hand,the large recommendation system with trust data faces the sparse trust problem,which seriously affects the reliability of trust propagation and creates a problem of grade inflation.In addition,cold start and sparse evaluation problems become more serious and frequent in large recommendation systems such as Taobao and eBay,which exacerbates the sparsity of trust and leads to a decline in recommendation accuracy.This paper summarize the concept of trust,sparse trust,trust field model and trust mining framework,which lays a foundation for trust-related research of large-scale recommendation system.Based on this,an anti-sparse trust model based on text trust source is proposed.The anti-sparse trust model in this paper is based on the text trust source to complete the modeling process.The sparse trust model first needs to preprocess the text comments,that is,to classify multiple sentences contained in the text;Then the semantic of each classification is computed,and the user's feature vectors are mapped according to the trust weights of each classification,and the user's similarity is calculated according to the feature vectors;Finally,according to the trust propagation mechanism,the sparse trust value is calculated by the sparse trust mining method,and the reliability of the sparse trust value is judged according to the user similarity calculated earlier,and the reliable trust data is filled with the matrix to achieve the purpose of anti-sparsification.This paper designs two experiments to prove the advancement of our algorithm.We conducted a horizontal experiment for the sparse trust model.The experiment selects a number of sparse matrices with different sparsity to verify the stability,update and effectiveness of the sparse trust method.In addition,we also selected three different trust models,running three different real big data sets in the same experimental environment for longitudinal comparison experiments,which proved that our method has better anti-sparsification ability than other methods.The de-sparse trust model based on text trust source is different from the traditional trust model.It fully exploits the user's real evaluation in text evaluation,and applies this model to the sparse problem that needs to be solved in the current big data environment.