Research On Social Recommendation Based On Trust

With the rapid development of Web2.0, the competition in electronic commerceplatform has been the focus in the present Internet world. Under this condition, therecommendation system attracts more and more attention from researchers. Therefore,a series of recommendation algorithm are proposed, such as classical collaborativefiltering algorithm, recommendation algorithm based on content-based,recommendation algorithm based on graph structure, and hybrid recommendationalgorithm. In this work, the trust relationship is introduced into the recommendationsystem based on the social network environment, which can achieve more accuraterecommendation results, which as specified bellow:Firstly, we successfully predicted the trust relationship strength between users insocial networks. We start the discussion from the attribute features of users since eachuser has the different attribute features. Then we analyze user’s attribute based onuser’s individual data quantitatively and combine all of the attributes into the fourattributes evidence. Quantitative attributes discreted as qualitative attributes throughdiscretization method, giving the qualitative analysis of attributes evidence. Next, webuild the attribute recognition framework based on qualitative attributes evidence anddistribute trust to evidence through membership degree principle. Then the model wasconstructed, achieving the prediction of trust relationship strength between users. Theresult of prediction is a triple rather than a traditional number.Then, social relationship is introduced into the recommendation system, achievingthe social recommendation. Firstly, we introduce the model of recommendationsystem visually, and describe all kinds of recommendation algorithms in detail, andfurther show the evaluation criteria of recommendation system at the same time. Then,in order to solve the cold start problem in the present recommendation algorithm, wedivide users into active and inactive, through this way the recommendation model isbuilt for the two kinds of users. In order to solve the sparsity problem of user-itemmatrix, we introduce trust relationship into recommendation system. The trustrelationship is the fusion result of combination of almost all of the user attributes, thusit has better authority and reliability than the method of building user-item matrix only based on similarity. It solves the sparsity problem effectively, and increases theaccuracy of prediction.Lastly, we conduct experiments on the prediction of trust relationship in socialnetworks and recommendation in social networks respectively. In the prediction oftrust relationship, we verify the sufficiency of the attribute evidence firstly, and then,the trust prediction result is also verified by the way of seven fold cross-validation andcontrast with other machine learning methods. These experiments proved thefeasibility and realizability of the trust combination model. In social recommendationexperiment, we conduct experiments to the recommendation algorithm of active usersand inactive users based on assessment criteria in recommendation system, incomparison with traditional collaborative filtering algorithm. These experiment resultsverified that the recommendation algorithm based on trust relationship proposed inthis paper has higher precision and recall.