Research On Review Spammer And Groups Detection Based On Deep Reinforcement Learning

Online product reviews influence consumers’ purchasing decisions on e-commerce websites.Some merchants might hire reviewers or a group of reviewers to promote or demote a set of target products by writing fake reviews.The flood of fake reviews has affected consumers’ shopping experience and the normal operation of e-commerce websites.To detect fake reviewers and fake reviewer groups,researchers propose solutions from different perspectives.These methods mainly have the following shortcomings: 1)There is no unified framework for detecting both fake reviewers and fake reviewer groups;2)Relying on hand-crafted features for detecting collusive spammers,which is costly and time-consuming;3)The quality of candidate groups directly affects the fake reviewer groups detection.Aiming at these problems,this paper proposes deep reinforcement learning-based methods for detecting fake reviewers and fake reviewer groups.Firstly,in view of the problems of no unified framework for detecting both fake reviewers and fake reviewer groups in existing detection methods,this paper proposes a unified framework for detecting fake reviewers and fake reviewer groups based on deep reinforcement learning.The framework unified modeling of fake reviewers and fake reviewer groups detection is as follows: constructed a collusive user relationship graph as the interaction environment.The fraud detection agent obtains the suspicious fraud degree of graph user nodes based on Q evaluation network and determines the user node space that can be selected according to the characteristics of fake reviewers user and group detection.Design different strategies for selecting suspicious fraud user nodes and provide environment reward feedback,continuously executing strategies to obtain suspicious fraudulent user node termination sequences and detect fraudulent users or groups.Secondly,aiming at the problem of hand-crafted features for detecting collusive spammers,this paper proposes fake reviewers detection method based on deep reinforcement learning based on the aforementioned framework.In this detection method,the fraud user detection agent excludes the already selected graph user nodes to determine the user node space that can be selected;executes the strategy of selecting suspicious fraudulent user nodes based on-greedy algorithm;and detects fraudulent users using the generated suspicious fraudulent user node termination sequence.The effectiveness and rationality of the detection method are verified by comparing it with existing methods.Finally,aiming at the problem of cannot achieve end-to-end fake reviewers group detection,this paper proposes fake reviewers group detection method based on deep reinforcement learning.In this detection method,the fake reviewers group detection agent,under the constraint of group size,determines whether the next node to be selected on the graph is the initial point of a group and determines the user node space that can be selected based on the node selection status.It executes a-greedy strategy for selecting suspicious fraudulent user nodes combined with priority nodes,and detects fake reviewers groups using the saved group initial points and the suspicious fraud user node termination sequence.The effectiveness and rationality of the detection method are verified by comparing it with existing methods.