چكيده به لاتين
Nowadays, social networks play a very important role in the exchange of information and communication among the people of the world. The increasing spread of social networks and its impact on various dimensions of life and human societies have, on the one hand, great benefits to human life, and on the other hand, has created a capacity for many security problems, such as subversive behavior, fraud, terrorist operations, spam, and so on. Identifying anomalous and destructive behaviors is an essential requirement. Anomaly in social networks are referred to as behaviors and patterns that are not consistent with the general behavior of those networks. The methods available to identify abnormal behaviors in social networks focus on a particular behavior of users, while human behavior and exchanges are inherently multiple and diverse. Multilayer networks are used to model such behaviors. In order to identify malicious people, in this research, a method based on the static communication graph and integration of information obtained from different layers in the social network is presented. The proposed method is an unsupervised method based on the structural characteristics of the network. In the proposed method, people's abnormalities are measured based on two star-near star and clique-near clique patterns. First, the collection of input data is categorized into different layers and then extracted in each layer for all existing nodes based on the patterns mentioned. In the next step, these scores are merged into different layers based on the weight of each layer and the ultimate anomaly score is obtained. The integration of the disparity scores from different layers is based on layer relevance for individual nodes, which is calculated for each node in each layer individually. We also integrated the method of local and non-local attitudes, in addition to considering the one-step neighbors of a person in computing, it considers two-steps neighbors. This makes it possible for those who interfere with other people to hide their behavior. The Noordin Top Terrorist Network and the Social Evolution Dataset have been used to evaluate the proposed method. The evaluations prove that the proposed method works more accurately than the methods available at the knowledge boundaries to find hidden patterns.
