Research On Collaborative Filtering Recommendation Algorithm Based On Trust Mechanism

With the rapid development of information technology and social networks,the massive data exploding at the exponential level has caused serious information overload problems.To solve this problem,the recommendation system helps users efficiently filter worthless information and proactively recommend personalized service information.In the recommendation system,the collaborative filtering recommendation algorithm is widely used due to its domain independence and the advantages of mining users' potential interests.However,it inevitably faces enormous challenges such as data sparsity,cold start issues,and malicious attacks.Therefore,to improve the recommendation reliability and effectiveness of the collaborative filtering recommendation system,this paper effectively combines the trust mechanism with the collaborative filtering recommendation algorithm by introducing the trust relationship among users in the social network.A collaborative filtering recommendation algorithm combining information entropy similarity and dynamic trust is proposed.The main research work is carried out from three aspects: improving the similarity algorithm,constructing the trust computing model,and introducing the trust reward and punishment mechanism:(1)For the problem that the accuracy of similarity caused by sparse user score data is low or the similarity of cold start users is difficult to calculate,construct the information similarity degree similarity calculation method based on score difference,and introduce an implicit similarity of trust relationships.Finally,the two similarities are adaptively dynamically fused to calculate the user's comprehensive similarity.The improved user similarity algorithm in this paper avoids the irrational phenomenon that the existing similarity algorithm is too high or too low due to the rare coevaluation project;When the user score data is sparse or not scored,the implicit similarity in trust is used to optimize the similarity calculation,which alleviates the cold start problem to some extent;(2)For the unreliable score data of malicious attack and false feedback in the recommendation system,the user's recommendation trust degree is measured from three dimensions: user score credibility,score prediction ability and co-evaluation item set.The explicit trust and recommendation trust are combined to calculate the direct trust degree between users.Based on the reconstructed trust network analysis,the single path trust weak propagation and multi-path trust aggregation mechanism are studied,and the indirect and global trust degree calculation model is constructed.In this paper,the direct trust degree fully reflects the recognition degree of the target user's recommendation or score to the recommended users,and mines recommended users that are trusted by the target user and have similar interests through trust reasoning.(3)In order to reflect the dynamic nature of the trust relationship between users,a trust reward and punishment update mechanism is introduced.By evaluating the effectiveness of the recommended user ratings,the trustworthiness of the active recommenders is reasonably rewarded,and the trustworthiness of the negative recommenders is reasonably penalized.A reasonable trust reward and punishment strategy can dynamically update the recommended users of the target users,thus characterizing the dynamic evolution of the trust relationship between users.The trust reward and punishment update mechanism proposed in this paper can effectively reduce the false recommendation of unreliable users,so that the recommendation result always comes from the recommended users who are trusted and like the target users and improve the recommendation reliability of the system.Finally,experiments were conducted based on the public trust data set FilmTrust?Epinions,and compared with other classical algorithms.The experimental results show that the proposed algorithm can significantly improve the recommendation accuracy of the recommendation system,and effectively alleviate the data sparseness,cold start and data reliability of the collaborative filtering recommendation system.