LEMD: Latent Environment Extrapolation and Message Disentanglement for Dynamic Graph Under Distribution Shift

Faithful, Sufficient and Understandable: Rethinking Graph Counterfactual Explanations via Discrete Diffusion Inversion

This work proposes Graph Diffusion Counterfactual Explanation via Inversion (GDCE-I), a discrete denoising diffusion model with a novel discrete inversion scheme that enables distribution-aware edits leveraging the whole domain edit space and qualitatively shows that GDCE-I attains interpretable in-distribution solutions.

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

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.

Disentangled Hypergraph Network with Implicit Structure Learning for Mobility Social Relationship Inference

A DHISL network that first captures individual spatiotemporal characteristics through feature-guided representation initialization, and then adopts a dual-branch framework that introduces dual constraints in a multi-channel disentangled space to achieve intra-channel consistency and inter-channel exclusivity.

Bridging the Gap Between Plausibility and Admissibility: Constraint-Aware Flow Maps for Dynamic Graph Systems

Robust Cross-Domain Sequential Recommendation Driven by Causal Disentanglement and Counterfactual Generation

Sequential recommendation systems are confronted with the dual challenges of data sparsity and domain bias. Especially in cross-domain scenarios, user interests are deeply coupled with domain-specific noise, which severely restricts recommendation performance. Targeted at improving the robustness of cross-domain sequential recommendation, this paper proposes a novel framework driven by causal disentanglement and counterfactual generation, namely Causal Disentanglement and Counterfactual Generation (CDCG). The framework focuses on three key components of sequential recommendation: representation learning, data augmentation and robust training. Firstly, the Cross-Domain Causal Decoupling (CDCD) based on front-door adjustment is adopted to separate genuine user interests from domain biases in an unsupervised manner, so as to provide purified causal representations for cross-domain recommendation. Secondly, guided by structural causal models, the framework generates Counterfactual Sequence Generation and Validation (CSGV) to augment training data while guaranteeing the rationality of temporal logic. Finally, by integrating distributionally robust optimization and mechanism preservation regularization, the Counterfactual Robust Optimization and Mechanism Preservation (CROMP) module enhances the model’s adaptive capability to leverage noisy counterfactual data. Extensive experiments on two benchmark datasets, Amazon and MovieLens, demonstrate that CDCG outperforms the state-of-the-art baselines by 11.1% in extremely sparse settings with only 1% of data available (p < 0.001). In particular, it achieves a 17.5% improvement for ultra cold-start users (1–2 interactions), who constitute 35% of the target population under 1% sparsity, with progressively smaller gains for users with 3–5 interactions (13.8%) and 6–10 interactions (10.1%). The results verify the effectiveness of the causal-driven paradigm in promoting the overall performance of cross-domain sequential recommendation systems.

Beyond the Bidirectional Promise: Re-evaluating the Robustness of Diffusion Language Models

While highly stochastic DLM loss landscapes naturally resist gradient-based adversarial suffixes, they provide no guaranteed defense against natural noise, proving that everyday robustness is weight-dependent rather than inherently architectural.