A Deep Graph Convolution Network-Based Abnormity Detection Model for Largescale Behavioral Data

• Published in: IEEE access Vol. 12; pp. 94380 - 94392
• Main Authors: Liang, Shaolin, Shao, Kangjie
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 2024; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Accuracy; Anomaly detection; Behavior; behavioral modeling; Behavioral sciences; Complex networks; Convolution; Convolutional neural networks; Data models; Deep learning; Feature extraction; Graph convolution network; Graph convolutional networks; Graphical representations; high-dimensional data; Internet; Performance evaluation; Social networks
• ISSN: 2169-3536; 2169-3536
• DOI: 10.1109/ACCESS.2024.3424879

It remains important to make abnormity detection from largescale behavioral data of Internet. Existing related approaches mostly failed to employ high-dimensional characteristics of Internet data, which limits the detection effect. To deal with this issue, we introduce graph convolution network (GCN) to generate fine-grained feature representation towards largescale behavioral data. And a deep GCN-based abnormity detection model for largescale behavioral data is proposed in this paper. Firstly, GCN is used to extract global co-occurrence information from largescale behavior data. Then, global embedding is applied to the encoder to obtain local features, which are fused into advanced features to better capture the relationships among nodes in social network. Finally, based on the idea of support vector domain description, a new objective function is optimized to determine whether abnormal behavior occurs in behavior data. Empirically, we have also carried out some experiments to make performance evaluation. The research results indicate that the proposal has higher Precision and robustness compared to traditional methods.