Complex Interaction Relations Based Anomalous Node Detection Algorithm In Social Networks

Anomaly detection has a wide range of applications in real life and has attracted the increasing attention of people.The complex interaction between multiple entities widely exist in social networks,which can reflect specific human behavior patterns.It’s helpful to detect anomalous node in social networks by paying attention to these complex interaction relations.However,due to the lack of an effective mechanism in most existing graph learning methods,these complex interaction relations fail to be applied in detecting anomalies,leading to inaccurate detection results.In order to address the aforementioned issue,this paper proposes two novel anomalous node detection algorithm in social networks.This paper presents a higher-order structure based anomalous node detection method(GUIDE)in social networks.It use the higher-order structures(i.e.,network motifs)to model complex interaction relations between multiple entities to promote anomalous node detection in the social network.Specifically,it exploits attribute autoencoder and structure autoencoder to reconstruct node attributes and higher-order structures,respectively.Moreover,a graph attention layer is designed to evaluate the significance of neighbors to nodes through their higher-order structure differences.Finally,this work leverage node attribute and higher-order structure reconstruction errors to find anomalies.Besides,this paper presents a multi-source data based anomalous node detection algorithm(NOTICE)for the problem of fake news detection in social networks.This method utilizes different types of network motifs to model complex interactions between news and other entities on heterogeneous networks,which is helpful for fake news detection.First,the motif-level embedding of a node is obtained by aggregating the network motif based neighbors of the target node,and then a hierarchical attention mechanism is designed to aggregate node features from both motif-level and semantic-level perspectives.The final aggregated node representation effectively captures the structural and semantic information in the network,so it can be used to better detect fake news.This paper conducts a lot of experiments on the two proposed algorithms.The experimental results show the superior performance of the two algorithms,and further verify the importance of complex interaction relations between multiple entities in social networks for anomalous node detection.