A Multi-Agent System Trust Model Based On Social Network And Reputation

Multi-agent system based e-commerce system has the great advantage to the traditional e-commerce technology in mobility, coordination and intelligence, and has tremendous development in recent years. In most of the multi-agent system, the nodes in the system choose the most suitable traders relying on the trust mechanism to avoid the chicane from the malicious nodes. Because the multi-agent system is relatively large, direct interaction record can't be found between the agents, and direct trust is often difficult to obtain. Therefore, the demand for the reliability and accuracy of recommended reputation value is increasing. Amounts of existing trust models don't consider the source of the information when calculate the recommended reputation value. Although they use the measures such as maintenance recommended reputation value to aggregate the recommended information, the update of recommended reputation value is always the technical difficulties and the mechanism is difficult to avoid the interference of malicious nodes.In order to get the recommended nodes more trustable, this paper uses the views of position structure and network structure in the concept of social network structure based on the topology of multi-agent system which can calculate the recommended reputation value more accurately to mine the social network. The recommended information provided by these nodes can be aggregated by distributing these nodes in different reputation values. Combined with the existing trust models, this paper proposes a multi-agent system trust model based on social network and reputation trust mechanism and specific researches include:First, this paper analyzes the topology of the multi-agent network and expresses it in a graphical way. Then, the trustable nodes would be found out by using the social network mining through the view of social network structure, and would be given different reputation value depended on different situation.Second, a reputation trust model based on social network is proposed by improving and optimizing the existing reputation trust model. This model considers the risk factors when calculates the direct trust value, and combines the nodes mined by social network when calculates the recommended reputation value, and proposes the weight between direct trust value and recommended trust value based on the history transaction amount when calculates the composite trust value. When calculating the global trust value, the threshold is set to increase the resistance for model to defend the attack from the oscillation nodes. Also, the simulation experiment of this model verifies the validity of this model.Finally, a prototype system is developed and realized under an application scenario based on the trust model proposed by this paper, and achieves the validation of the model in feasibility and effectiveness. The result shows that the model proposed by this paper has some reference value to effectively judge the trust value of the nodes and find out the malicious nodes.