Study On Trust Modeling And Trustworthy Recommendation Methods On O2O Service

With the rapid development of mobile communication technology and mobile computing technology, the O2O E-commerce is widely accepted by the public, it makes people' life easier than ever, and has changed their life style. However, most of O2O service providers are little local service providers, lack of scientific reasonable management methods, and they can hardly be supervised by government. So, there are many problems occurred in this hot phenomenon, some O2O service providers provide poor services to their users, some O2O service providers are dishonest, and so forth.These problems have made people economic losses, and have done harm to the health of O2O E-commerce.Aiming to solve these problems, this paper analysis the trading features of trading entity in O2O E-commerce, to constructing the reputation model of O2O service providers, so prevent them to do dishonest actions such as collusion, fraud. Meanwhile,we build the trust network based on target user's social network, to identify the reliable neighbors which can give reliable advices to target users when they consulting for suitable services or suitable service providers. After that, we take target users' personal preference into consideration, to recommend them with trustworthy services and trustworthy service providers. Furthermore, we introduce the SEMBA model which derives from the Advogato trust model into our recommendation methods, so as to reduce the running cost of our recommendation methods, and become more suitable for mobile context. The contributions of this dissertation are fourfold:(1) To identifying the trustworthy group in personal mobile social networks effectively, this paper proposes a trust model based on the famous Advogato model, it takes the interaction frequency, similarity of users' social circles, similarity of users'interests into consideration, uses the capacity-first maximum flow search method to spread the trust flow between users to build their personal trust networks, and finally output the ranked trustworthy user group. The experimental results on real dataset show that, our trust model has better performances on the prediction accuracy(Pre), missing rate( MsR), and Top ranking range( Trr ) while comparing to the existing related models.(2) We propose a lightweight trustworthy service recommendation method based on the personal local social network. This method uses a modified TMBA trust model to build the local trust network of the target user firstly, then to compute the local trust weights of users based on the similarity of social circles and interactions of users, and finally to do the service rating prediction of target users based on the local trust weights and the user rating similarity which is computed based on the user-service rating matrix.The comparing experimental results show that our recommendation method has better service rating prediction accuracy, has lower computing complexity, and can count against the Shilling attacks well while comparing to other state-of-art service recommendation methods.(3) We propose a novel reputation computing model ESRep on O2O E-commerce sellers. Besides the subjective factors such ratings from customers, transaction price deviation degree, the objective factors such as seller's running time, passenger flow are also took into consideration in this model, so it can overcome the subjectivity of traditional reputation models, can reflects the real reputation of offline service providers of O2O E-commerce. Simulation results show this model can effectively reflects the real reputation of O2O E-commerce seller, reduces the effect of hostile nodes' rating,and resists the reputation collusion action from hostile nodes.(4) We propose a trustworthy recommendation method to recommend suitable O2O service providers to users based on providers' reputation and users' similarity. This method uses the user-service rating matrix to computing users' comprehensive ratings on O2O service providers, and generates the comprehensive user-service provider rating matrix. Based on this matrix, our recommendation method combines it with the reputations of O2O service providers to compute the similarities of different users, then to recommend suitable O2O service providers to users. Simulation and experiment results demonstrate the proposed method has better recommendation accuracy while compare to the other traditional recommendation methods and start-of-art service recommendation methods, but also it has better performance while counter against malicious cheating actions.