Skip to content
Log in
Book

Rethinking Generalization in Graphs: A Hierarchical Interaction Perspective for Generalist Detection

Xiangping Zheng Xuan Feng Bo Wu Li Yang Bin Ren Wei Li Bin Tang Zhiwen Yu
Aug 2026 · Proceedings of the 32nd ACM SIGKDD Conference on Knowledge Discovery and Data Mining V.2 · 0 citations · 10 references

TL;DR

Hierarchical Interaction MOdeling for zero-shot generalist GAD enables anomaly detection across diverse graph domains without retraining or access to target-domain supervision by modeling the evolutionary trajectories of node representations across hierarchical structural depths, thereby capturing interaction patterns that exhibit strong cross-domain stability.

Abstract

With the increasing heterogeneity of social networks and online interaction systems, generalist graph anomaly detection (GAD) has become essential for identifying abnormal and fraudulent behaviors in complex environments. However, most existing GAD approaches rely heavily on domain-specific semantic alignment, which substantially restricts their ability to learn transferable node representations and often leads to poor generalization on unseen graph domains. To address this challenge, we propose HIerarchical Interaction MOdeling for zero-shot generalist GAD (termed HIMO-GAD). HIMO-GAD enables anomaly detection across diverse graph domains without retraining or access to target-domain supervision by modeling the evolutionary trajectories of node representations across hierarchical structural depths, thereby capturing interaction patterns that exhibit strong cross-domain stability. Specifically, HIMO-GAD integrates two core components: (1) a Dynamic Interaction Modeling Module that characterizes cross-layer interaction evolution to extract transferable representations, and (2) an Anomaly-Aware Regulation Mechanism that combines gradient immunity and centralization regularization to suppress overfitting and stabilize cross-domain generalization. Extensive experiments on multiple real-world graph datasets demonstrate that HIMO-GAD consistently outperforms state-of-the-art baselines in strict zero-shot settings, achieving up to a 10% improvement in key evaluation metrics and exhibiting strong generalization across heterogeneous graph domains.

DOI Semantic Scholar
View source

Similar papers

Open access Aug 2026

Hybrid-HGT: A unified deep learning framework for community detection in heterogeneous attributed graphs

A combined HGT-based framework incorporating contrastive representation learning and deep clustering with multi-round training via pseudo-labels is presented, enabling the model to better deal with label scarcity, heterogeneous dependencies, and overlapping semantics in practical, complex, attributed networks.

Hamza Haddad, Hicham Attariuas, A. Younes · 0 citations
Book Open access Jul 2026

SNBot: Modeling Self–Neighborhood Representation Discrepancy for Social Bot Detection

SNBot is proposed, a novel social bot detection framework that explicitly models the discrepancy between node self-representations and their neighborhood embeddings and consistently outperforms state-of-the-art methods.

Qilong Lin, Jingya Zhou · 0 citations
Jul 2026

Anomaly Detection via Contrastive Learning on Graph using Embedding

A graph embedding-based error detection method via contrastive learning that treats each triple as a hyper-node, enabling data-driven modeling of semantic consistency and integrates a semantic matching-based scoring mechanism with a contrastive learning-based scoring mechanism.

Gaojun Shan, Liping Cong, Xiaohong Sun et al. · 0 citations
Open access 2026

GCTMR: Graph Contrastive Transformer for Multibehavior Recommendation

Multibehavior recommendation aims to leverage diverse user interaction types—such as views, clicks, and purchases—to better understand user intent and enhance recommendation accuracy. However, effectively modeling these heterogeneous behaviors remains challenging due to the sparsity, noise, over-smooth embedding, and dynamic nature of interaction sequences. To address these issues, we propose GCTMR (Graph Contrastive Transformer for Multibehavior Recommendation), a framework that integrates graph-based contrastive learning with a behavior-aware Transformer architecture. Specifically, the methodological contribution lies in three key innovations: 1) an adaptive sparse attention mechanism employing $\alpha $ -entmax that dynamically filters out noisy or irrelevant interactions to preserve the distinctiveness of user behavior representations; 2) a frequency-aware multibehavior encoder using FFT/IFFT decomposition that disentangles stable long-term preferences from short-term behavioral shifts across different interaction types; and 3) a simplified focal loss function that emphasizes hard-to-predict instances, improving robustness under data sparsity and behavioral ambiguity. The empirical validation is exceptionally strong, demonstrating improving of 22.43% on Tmall and 18.10% on Beibei over state-of-the-art baselines in HR@10 metrics, with particularly remarkable performance under sparse data conditions where traditional methods struggle most. Comprehensive ablation studies confirm that each component contributes meaningfully to the overall performance, while extensive hyperparameter analyses demonstrate the framework’s stability and practical applicability.

Nikorn Kannikaklang, S. Wongthanavasu · 0 citations