A Dynamic Trust Evaluation Model of User Behavior Based on Transformer

In the boundary-based protection system, although user authentication can provide a certain degree of security, when user information is leaked, this method will be difficult to deal with the attacking threat from internal and external legitimate users. Aiming at the attacking threat of legitimate users, we analyze the user behavior and constructs a dynamic trust evaluation model of user behavior based on the Transformer network. Firstly, unsupervised pre-training of the Transformer network and extracting the time characteristics of data can effectively improve the model’s generalization ability. Then, fine-tune the network parameters, build the trust model of users’ historical behavior, predict a user’s future behavior through the established trust model, and calculate the similarity between the predicted behavior and users’ actual behavior to evaluate the trust of users’ behavior. Finally, we design and conduct experiments on a public data set. The experimental results prove the effectiveness of this method and also show that this method can reduce the training time and improve the accuracy of trust evaluation. © 2022, International Journal of Network Security. All Rights Reserved.