Comparative Analysis of Graphical Structural Metrics for Identifying Anomalies in Online Social Networks

Social networks are exposed to a variety of security problems due to their wide use and popularity. Therefore, identifying unusual activities in social networks is of paramount importance as it helps to obtain significant information about the behavior of unusual users and identify them. One of the important aspects of social network analysis is to check the presence of anomalies. Anomalies in the field of social networks imply irregular and often illegal behavior. A host of methods have been proposed to detect different kinds of anomalies in social networks. According to the employed approach, these methods can be classified into three categories, namely, clustering-based, based on network structure-based, and signal processing-based. In this paper, we extend the graph structure-based approach by introducing and analyzing important graph metrics to detect abnormal activities. Theoretical and experimental evaluation using several large data sets demonstrate that the relationship between the interface node and the number of edges helps to correctly detect and rank the maximum number of anomalies.

The proposed method is a combination of graphical and statistical theory. First, various metrics and graph structures are calculated, and then statistical methods are used to identify and analyze unusual structures (stars and clusters).

Statistical and visual analysis shows that the area covered by the curve is maximum for the interface (B) compared to the number of edges (E). The results show that the proxy is a scale that can correctly detect many abnormalities. It can also be said that the relationship between the (B) and the (E) helps to predict most anomalies.

In this research, a structure-based method was presented by using graph criteria to predict abnormalities. The curve fitting method based on the graph structure was extended to detect anomalies using the combination of new graph criteria. It was observed that the relationship between the interface and the edges helped to predict a large number of anomalies that were either misclassified or missed by the Oddball method and the ABC relationship to E. The abnormality scores assigned to the nodes help predict the degree of anomalies and rank the nodes according to their irrational behavior.