Reasearch On Abnormal User Identification Based On Target Terminal And Social Data

With the rapid development of Internet technology,information systems bring convenience to people's lives,but also provide many illegal users with the opportunity to attack from the perspective of social engineering,and how to protect the privacy and security of users has become an important research topic.From the perspective of social engineering,this paper focuses on two attack methods: impersonation(to achieve one's own purpose by impersonating the identity of others)and social relationship(to attack by using a certain connection between users).Protecting user privacy security involves personal terminal information protection and user information security based on virtual network,and abnormal user detection is an important part of protecting user privacy security and an important step of maintaining information security,because most abnormal users are attackers.Therefore,this study from two scenarios based on terminal data and social data to detect abnormal users.Firstly,to solve the problem of low recognition rate of single key features and prevent impostors from restricting an input device to avoid detection,this paper proposes a method of user identity anomaly recognition based on Multiple Kernel Learning for fusing mouse and keyboard feature.Due to the differences between keyboard features and mouse features,this paper proposes to map each type of feature to a suitable kernel function for nonlinear mapping,and then get the weight of each kernel by calculating,and get a combined kernel by summing,so as to realize multi-feature fusion.The data set used in this paper is to collect real user's mouse and keyboard data in uncontrollable environment through the acquisition program.The experimental results show that the effect of the traditional method is not ideal,and the proposed method gets more stable and effective identification than single index and double index feature recognition.Mouse and keyboard are the input devices in the hardware environment without network.In the real virtual network,because the advanced abnormal users who can imitate human behavior need to keep in touch with other users in order to achieve certain communication purposes,the topology of social network is stable in a certain period of time.Therefore,this paper proposes a social robot detection method based on network topology,and uses deep learning method to mine hidden complex information.In this paper,we choose graph embedding method and improve algorithm to mine deep local features,and then fuse the feature matrix and important attributes of nodes generated by graph embedding as the input of graph convolution neural network based on graph filtering,because using graph convolution neural network detection based on graph filtering can not only obtain global features,but also increase the learning rate of a small number of labels,so as to improve the detection rate.