Research On Trust Model Based On Multi-attribute Ratings

With the rapid development of internet technology,more and more web applications appear in people's daily lives,such as P2 P network,social network,multi-agent system,electronic commerce.However,in the expansive web application which is dynamic and openness,a great number of malicious users have appeared,and caused various degrees of losses to people.In order to reduce people's losses and satisfy their demands,more and more trust models were proposed by researchers to help people make effective decisions.In this thesis,we comprehensively analyze the advantages and disadvantages of the current classic trust models,and then propose a new multi-attribute rating based trust model based on these references and the definition of trust.In this model,the calculation of comprehensive trust consists of direct trust and indirect trust.For direct trust,it is based on direct interaction ratings of providers.Moreover,since the behaviors of providers are dynamic and the previous experiences can't determine the future performance,we make use of a dynamic forgetting factor to discount the older ratings.For indirect trust value,it was calculated according to other users' recommendation.Firstly,we compare the user's ratings with the recommender's ratings for commonly rated providers to get the rating pairs.Secondly,we calculate the similarity of rating pairs to acquire recommend trustworthiness by using Euclidean norm,and adjust it through a probability-based certainty measuring method.Thirdly,we take advantage of the transformation from the Dempster-Shafer theory to the beta function to get indirect trust.After get the aforementioned results,the comprehensive trust was modeled by combining the weighted of direct trust and indirect trust.In addition,we have considered the useful information of malicious users' rating to provide various evidence sources for trust model.Different with traditional single rating based trust models,our proposed model take full use of the multi-attribute rating and users' preferences to evaluate the trust value of provider.By comparing with other classic trust model in the experiments,it demonstrates that our model is more effective in resisting different types of attacks.Therefore,the research of this model has a certain theoretical significance and application value.