A Reputation-Based Trust Evaluation Model for P2P E-Commerce

Abstract

P2P e-commerce as a new shopping pattern is developing rapidly, and users can transact with anybody anytime and anywhere in the P2P e-commerce system, but as users often interact with strangers, hence building trust among the users, make the users know each other well, estimate the risk of participating in a transaction, and enable to transact securely in the P2P e-commerce system is an urgent problem to be solved for P2P e-commerce nowadays. In order to settle the problem, this paper provides a novel reputation-based trust model in P2P e-commerce system. The main thought of the proposed trust model is that a node has two types of reputation: local reputation and global reputation. In the local reputation computing, we take the calculation of factors including the amount of transaction money and transaction time and so on, and in the global reputation computing, there are three factors introduced—the number of neighbor nodes, the direct trust level of neighbor nodes, and the trust evaluation weight of neighbor nodes, and finally the object node's trust level can be obtained by using the local reputation weight coefficient and the global reputation weight coefficient to integrate the node's local reputation with global reputation. We have conducted two experiments that show that our trust model contrast with the present typical trust models EigenRep and Beta in the aspects of the computing load and the successful transaction rate. The first experiment shows that the trust model has a lower computing load compared to the EigenRep and Beta, because we do not compute the node's reputation through the transitive recommending of the intermediate nodes; the second experiment shows that the trust model has a higher successful transaction rate when the malicious nodes exist in the system, and even the number of malicious nodes proportion is larger, the trust model also has a higher successful transaction rate. In conclusion, the trust model more roundly takes into account the factors affecting the local reputation and global reputation and can greatly improve their computing methods. The experiment results show that the trust model can reduce the trust level computing load, and have more improvements on the successful transaction rate and anti-attack capability compared to the existing trust models. But the trust model does not introduce the risk factor, and the logic of the trust mode is relatively simple, so our trust model still needs to be improved constantly.