The Detection of Anomalous Users in Location-Based Social Networks by Using Graph Rules

An analysis of social networks is necessary to detect anomalous users, due to the popularity of these networks. This paper aims to detect anomalous users in location based social networks. For this purpose, an ego graph is computed for each user, and the five variables vertex degree, edge size, edge weight, betweenness centrality and eigenvector centrality are calculated with respect to the weights of the edges in this graph. Then six relationships between two of these variables are made up, and for each of these relationships, the line equation is obtained in the coordinate system of the two variables. This equation is used to predict the value of the variables. Based on this predicted value, the user's score is determined, and anomalous users are detected. The proposed method investigates anomalies in the friendship graph, location of residence and interests of users. The results indicate that the proposed method has been able to detect anomalous users by examining the scores of star and clique structures in the graph.