Search Results (2)

Graph anomaly detection (GAD) aims to identify nodes or edges that deviate from normal patterns. However, the presence of heterophilic edges in graphs leads to feature over-smoothing issues. To overcome this limitation, this paper proposes the multi-view heterogeneity resistant network (MV-GHRN) model, which progressively purifies heterophilic edges through multi-view collaboration. First, to address the noise sensitivity of single predictions, the method computes post-aggregation (PA) scores for both the original graph and its perturbed versions and performs weighted fusion, leveraging the consistency of multiple prediction perspectives to enhance the reliability of heterophilic edge identification. Second, a cosine similarity view is introduced as a complementary structural perspective, with both views independently completing heterophilic edge pruning to clean the graph structure from both topological and feature dimensions. Finally, a cross-view self-distillation mechanism is designed, using the fused predictions from the two purified views as teacher signals to guide the optimization of each view in reverse, correcting feature biases caused by heterophilic edges. Experiments on benchmark datasets such as YelpChi and Amazon demonstrate that the framework significantly outperforms existing methods. For instance, on the YelpChi dataset, MV-GHRN surpasses the best baseline by 16.8% and 5.2% in F1-Macro and AUC, respectively, validating the effectiveness of the progressive multi-view purification mechanism. Full article

Heterogeneous graph neural networks (HGNNs) deliver the powerful capability to model many complex systems in real-world scenarios by embedding rich structural and semantic information of a heterogeneous graph into low-dimensional representations. However, existing HGNNs encounter great difficulty in balancing the ability to avoid artificial metapaths with resisting structural and informational noise in a heterogeneous graph. In this paper, we propose a novel framework called Heterogeneous Graph Purification Network (HGPN) which aims to solve such dilemma by adaptively purifying the noisy heterogeneity. Specifically, instead of relying on artificial metapaths, HGPN models heterogeneity by subgraph decomposition and adopts inter-subgraph and intra-subgraph aggregation methods. HGPN can learn to purify noisy edges based on semantic information with a parallel heterogeneous structure purification mechanism. Besides, we design a neighborhood-related dynamic residual update method, a type-specific normalization module and cluster-aware loss to help all types of node achieve high-quality representations and maintain feature distribution while preventing feature over-mixing problems. Extensive experiments are conducted on four common heterogeneous graph datasets, and results show that our approach outperforms all existing methods and achieves state-of-the-art performances consistently among all the datasets. Full article