Research On Recommendation Algorithm Based On Trust And Fake Reviews Detection

With the rapid development of Internet technology,the explosive growth of Internet information has led to information overload problems.The recommendation system is an important tool for solving information overload.Collaborative filtering recommendations in personalized recommendation systems are by far the most widely used recommendation technology.However,with the wide application of the recommendation system,the recommendation algorithm has serious data sparseness,cold start and data security problems.The recommendation algorithm itself cannot effectively remove malicious data.To effectively solve these problems,this thesis deals with the false information in the recommendation system and studies the collaborative filtering recommendation algorithm based on trust and matrix decomposition techniques.The main work and innovations of this thesis are as follows:With the rapid development of e-commerce,there is a large amount of false information in the platform,which seriously interferes with the recommendation system of the platform,which makes the project satisfaction recommended to users decrease.This thesis proposes a fake review detection method that combines emotional polarity and D-S theory.The multi-dimensional feature model is constructed by considering the commentator's behavior characteristics,commenting text features and the emotional polarity of the review text.The preliminary recognition results of the support vector machine model under single feature are used as independent evidence to calculate the basic trust function,and the D-S evidence theory is used to conduct decision-level fusion and the comments are detected according to the evidence support degree under the recognition framework.Crawl data from the popular platform as a data set for experimentation.In the algorithm comparison,the proposed algorithm outperforms the detection algorithm proposed in the related literature.In the web2.0 era,the popularity of social networks has attracted the attention of researchers based on social network recommendation algorithms.In view of the current incomplete trust relationship,this thesis proposes a socialized hybrid recommendation algorithm based on integrated trust.The algorithm comprehensively considers the user's comprehensive trust and builds a trust matrix.And considering the degree of relevance between projects,combined with the probability matrix decomposition model to achieve collaborative filtering recommendation algorithm.The algorithm has achieved good results on the Epinions dataset.Then,once again,the data processed by the fake review detection technology is first applied to the system based on the integrated trust recommendation algorithm,and compared with the unprocessed data recommendation.Test the robustness of the proposed algorithm and whether the recommended effect is improved.Experiments on the Epinions dataset show that the recommendation score error and algorithm robustness are improved by the data recommendation based trust recommendation algorithm.