spatio-temporal graph modeling

Traditional ST-GCNs employ单一 temporal modules—either CNNs or LSTMs—leading to insufficient capture of dynamic spatiotemporal patterns. To address this, we propose a plug-and-play hybrid temporal module that, for the first time, synergistically integrates CNNs and LSTMs within a unified co-temporal block. This design jointly models local temporal features and long-range dependencies. Through theoretical analysis and cross-dataset ablation studies, we systematically characterize the intrinsic relationship between temporal module architecture and representational capacity. Evaluated on standard spatiotemporal graph benchmarks—including NTU-RGB+D and PeMSD7—our method achieves significant improvements in prediction accuracy and cross-domain generalization. It consistently outperforms pure-CNN and pure-LSTM baselines in temporal representation learning. The proposed module establishes a reusable, principled design paradigm for temporal modeling in ST-GCNs, advancing both expressiveness and architectural flexibility.

Current spatiotemporal knowledge graph (STKG) models suffer from fundamental limitations—including conceptual fragmentation, terminological inconsistency, poor reusability, and inadequate support for long-term knowledge preservation—arising from their disparate foundations in static, temporal, and spatial graph paradigms. To address these issues, this work introduces the first multidimensional analytical framework encompassing edge semantics, spatiotemporal annotation, and semantic modeling. Through a systematic literature review and cross-dimensional comparative analysis, we clarify the theoretical evolution of STKGs. We further propose a general-purpose modeling guideline explicitly designed for long-term knowledge preservation, establish standardized design principles, and distill six key open challenges. Our contributions provide both theoretical foundations and practical pathways for transitioning STKGs from application-specific solutions toward universal, sustainable knowledge infrastructure.

Existing spatiotemporal graph models—such as GNNs and Transformers—typically adopt decoupled spatial and temporal modeling, limiting their ability to capture nontrivial structural dependencies inherent in graph topologies. To address this, we propose Cy2Mixer, a three-module architecture integrating temporal modeling, standard message passing, and a novel recurrent message-passing block (RMPB). The RMPB’s core innovation lies in explicitly constructing and aggregating cycle subgraphs to encode topological invariants; we theoretically prove its information complementarity with conventional message passing, thereby overcoming the limitations of spatiotemporal decoupling. Additionally, Cy2Mixer incorporates a gMLP backbone, gating mechanisms, and mathematically grounded topological representations. Extensive experiments on multiple spatiotemporal forecasting benchmarks demonstrate state-of-the-art performance, with significant improvements in prediction accuracy and generalization—particularly for traffic flow forecasting.

Existing graph deep learning approaches for multivariate time series forecasting predominantly focus on architectural innovations, lacking systematic methodology studies on problem formalization, model design principles, and evaluation paradigms. This paper introduces the first unified methodology framework for graph-enhanced time series forecasting. It formally defines the forecasting problem by modeling dynamic inter-variable dependencies via learnable graph structures. It establishes principled design guidelines integrating graph neural networks (GNNs), spatiotemporal modeling, and deep temporal models (e.g., TCN and Informer variants). Furthermore, it proposes a reproducible evaluation protocol with built-in mechanisms to ensure interpretability and scalability. The framework shifts forecasting model development from empiricism toward first-principles reasoning, providing standardized guidance for relation-aware global modeling and articulating key open challenges. (149 words)

This paper addresses the problem of learning time-varying graph structures from sparse spatiotemporal measurements, where the core challenge lies in ensuring smooth and interpretable structural evolution under limited sample size. To this end, we propose a convex optimization–based framework for time-varying graph learning—the first to jointly impose two convex regularizers: (i) a graph Laplacian sparsity constraint and (ii) an ℓ₁-norm penalty on temporal differences between consecutive graph Laplacians, explicitly encoding temporal smoothness priors. The method integrates principles from graph signal processing with scalable iterative algorithms, guaranteeing both theoretical tractability and computational efficiency. Extensive experiments on synthetic data, real-world point cloud sequences, and temperature time series demonstrate that our approach significantly outperforms state-of-the-art methods—particularly in low-sample regimes—while exhibiting superior estimation accuracy, robustness to noise, and generalization capability.

Current temporal graph learning models suffer from unreliable evaluation and exhibit unexpectedly strong performance against simple baselines—a paradox suggesting overreliance on spurious or superficial graph properties. Method: We propose the first interpretable analytical framework grounded in eight fundamental structural and temporal attributes (e.g., density, recency, homophily), validated via controlled synthetic and real-world datasets. Using ablation studies and attribution analysis, we quantitatively assess how well seven state-of-the-art models capture each attribute. Contribution/Results: We find models effectively learn local connectivity but consistently underperform in modeling temporal recency and higher-order temporal homophily—revealing critical architectural limitations. Our work shifts evaluation from opaque end-to-end accuracy toward attribute-level interpretability, establishing a new benchmark for model diagnosis and trustworthy deployment of temporal graph neural networks.

Existing continuous-time dynamic graph methods struggle to capture long-range spatiotemporal dependencies due to their reliance on local neighborhoods. This work proposes CTDG-SSM, which introduces the first topology-aware memory mechanism by extending HiPPO through a Continuous-Time Topology-aware High-order Polynomial Projection Operator (CTT-HiPPO). This operator jointly models graph structure and temporal dynamics, yielding an efficient state-space representation. By integrating graph Laplacian polynomial projections with zero-order hold discretization, the framework achieves parameter-efficient modeling of long-range spatiotemporal interactions. Empirical results demonstrate state-of-the-art performance across dynamic link prediction, node classification, and sequence classification tasks, with particularly pronounced gains in scenarios requiring long-range reasoning.

Existing graph neural networks are inherently limited to modeling pairwise relationships, struggling to effectively capture higher-order topological structures while suffering from rapidly escalating computational complexity as graph size grows. To address these limitations, this work proposes a simplicial complex–based spatiotemporal neural network that, for the first time, integrates simplicial complexes into spatiotemporal modeling. By leveraging spatiotemporal random walks on high-dimensional simplicial complexes and parallelized temporal convolutions, the proposed method transcends the pairwise interaction constraints of conventional graph neural networks. This approach significantly enhances the capacity to model higher-order topological dependencies in complex systems while maintaining computational efficiency.

Existing heterogeneous temporal graph neural networks (HTGNNs) commonly adopt a decoupled spatiotemporal modeling paradigm, resulting in weak spatiotemporal interaction and high model complexity. To address this, we propose SE-HTGNN—a lightweight, end-to-end framework that tightly integrates spatial and temporal information. First, we design a history-guided dynamic attention mechanism that explicitly encodes temporal dependencies within spatial message passing. Second, we incorporate large language models (LLMs) via prompt learning to extract semantic priors of node types, thereby enhancing structural awareness. SE-HTGNN achieves state-of-the-art prediction accuracy while accelerating inference by up to 10× and significantly reducing computational overhead. Its core innovations lie in (i) internalizing temporal modeling into the spatial learning process, and (ii) the first systematic integration of LLM-derived semantic priors into heterogeneous temporal graph representation learning.

This work addresses the performance degradation of graph-structured multivariate time series forecasting under data-scarce and cross-domain transfer scenarios by proposing a structure-aware context selection mechanism within a transfer-oriented spatiotemporal graph learning framework. The approach incorporates an explicit graph context pruning strategy as an inductive bias, leveraging information-theoretic measures and correlation-based criteria to select informative subgraphs and features. This selection module is seamlessly integrated into a spatiotemporal convolutional architecture, enhancing both sample efficiency and out-of-distribution generalization. Evaluated under low-data transfer settings on large-scale traffic benchmark datasets, the proposed model significantly outperforms existing baseline methods.