Deep joint adversarial learning for anomaly detection on attribute networks

Haoyi Fan, Ruidong Wang, Xunhua Huang, Fengbin Zhang, Zuoyong Li, Shimei Su

Abstract: Highlights•We introduce JAANE, a framework for attribute network anomaly detection. It jointly learns attribute and structure embeddings, fusing them into a shared latent space. JAANE then identifies abnormal nodes using a compact hypersphere of normals.•We design a feature fusion module to capture attribute-structural consistency and complementarity. It employs multiplication-based fusion for attribute-network alignment and addition-based fusion for capturing attribute-structural complementarity.•Experimental evaluations on real-world attribute network datasets demonstrate JAANE's superior performance in anomaly detection compared to state-of-the-art methods.