Research On Trust Inference Algorithm Based On Double DQN

The development of social networks facilitates people’s daily communication and promotes online communication between users,but it also brings some risks to people.Trust plays an important role in social activities and can effectively avoid some risks from unreliable users.Because most users in social networks have no direct trust relationship,there are more indirectly connected through common friends.Therefore,how to build the trust relationship between indirect users and build a trust network has become a challenge.Indirect trust relationships can be used for trust inference through trust transitivity.Trust propagation and trust aggregation in the process of trust transitivity affect the results of trust evaluation to a great extent.How to find reliable trust paths and use effective aggregation strategies has become a major challenge.Due to the dynamic nature of trust,the trust value will change with the interaction between users.Good interaction experience will increase the trust value between two users,and vice versa.Because the trust propagation in social networks is very time-consuming,how to infer the dynamic trust value of two users who are not directly connected in a continuous time has become an important challenge.Reinforcement learning is suitable for multi-step decision-making.Previous studies have shown that applying reinforcement learning to trust inference can improve the accuracy of trust prediction.In response to the above problems,this paper studies the trust inference based on reinforcement learning of Double DQN: in order to predict the trust relationship between two indirect users,how users conduct trust propagation and trust aggregation in static networks is considered,a static trust inference algorithm is proposed.Firstly,a trust inference model is established based on Double DQN;secondly,the model is used to search the trust path,and the propagation function is used to evaluate the reliability of the trust path;finally,the interest similarity between users is considered,a new aggregation strategy is proposed.This aggregation strategy is used to aggregate multiple trust paths to evaluate the final trust value.The experimental results show that the prediction results of the proposed aggregation strategy are better than the existing aggregation strategies,and the proposed algorithm has higher prediction accuracy and precision than the existing trust inference algorithms.The dynamic changes of user trust value and the dynamic interaction between users are considered,a dynamic trust inference algorithm is proposed.Firstly,a two-layer neural network is used to simulate the dynamic interaction process of users and a reliable trust path is found;secondly,the trust value between users is dynamically updated according to the interaction results between users;finally,the proposed aggregation strategy is used to aggregate multiple trust paths to predict indirect dynamic trust.Experimental results show that the prediction accuracy of the proposed dynamic trust inference algorithm is better than other algorithms.The proposed algorithm is applied to the larger Epinions dataset,and the experimental results show that the proposed algorithm has good scalability.