Prediction and Abnormal Behavior Detection of Agent Dynamic Interaction Trust

Computation of trust is an interesting research direction in agent and multi-agent systems theory. There have been a lot of researches on agent trust and reputation in the past few years, such as TRAVOS model presented by Teacy and FIRE model presented by Huynh. However, previous work is only based on the average probability of historical interaction and there is a lack of attention to dynamic variety of agent trust. So the ability of precise prediction and abnormal behavior detection of trust are not satisfied. Due to these problems, using the probability theory, a computational model of agent interaction trust (CMAIT) where historical interaction is divided by time is proposed. Furthermore, based on derivative of trust, the computational confidence of computational information and computational confidence of deviation of CMAIT are given. The mechanism of detection of abnormal behavior of CMAIT is also given. It is proven that trust of average historical interaction is a particular case of CMAIT. Experiments are conducted on Web e-commerce at Taobao website. Experimental results demonstrate that computational error of CMAIT is half of that of TRAVOS and its computational complexity is also low. It can be applied in the detection of abnormal behavior and the prediction of future behavior. It improves the work of Jennings on agent trust.