Improving large language models (LLMs) for electronic health record (EHR) reasoning is essential for enabling accurate and generalizable clinical predictions. While LLMs excel at medical text understanding, they underperform on EHR-based prediction tasks due to challenges in modeling temporally structured, high-dimensional data. Existing approaches often rely on hybrid paradigms, where LLMs serve merely as frozen prior retrievers while downstream deep learning (DL) models handle prediction, failing to improve the LLM’s intrinsic reasoning capacity and inheriting the generalization limitations of DL models. To this end, we propose EAG-RL, a novel two-stage training framework designed to intrinsically enhance LLMs’ EHR reasoning ability through expert attention guidance, where expert EHR models refer to task-specific DL models trained on EHR data. Concretely, EAG-RL first constructs high-quality, stepwise reasoning trajectories using expert-guided Monte Carlo Tree Search to effectively initialize the LLM’s policy. Then, EAG-RL further optimizes the policy via reinforcement learning by aligning the LLM’s attention with clinically salient features identified by expert EHR models. Extensive experiments on two real-world EHR datasets show that EAG-RL improves the intrinsic EHR reasoning ability of LLMs by an average of 14.62%, while also enhancing robustness to feature perturbations and generalization to unseen clinical domains. These results demonstrate the practical potential of EAG-RL for real-world deployment in clinical prediction tasks.

Spatiotemporal forecasting is a fundamental task in areas such as traffic flow prediction, environmental sensing, and urban planning. Recent advances have shown that decomposing temporal signals into multiple frequencies and modeling them jointly with spatial structures can significantly enhance forecasting performance. However, existing multifrequency forecasting models still face two critical limitations. First, the importance of different temporal frequencies evolves over time, yet most models assume fixed or static frequency contributions. Second, spatial dependencies are inherently frequency-sensitive. For instance, low-frequency components often align with global spatial patterns, while highfrequency components tend to correspond to localized interactions. However, current approaches typically use a shared spatial information across all frequencies, introducing spatiotemporal inconsistency. To address these challenges, we propose a novel Adaptive Frequency Pathways (AdaFre) for spatiotemporal forecasting, which adaptively captures both dynamic frequency relevance and frequency-aligned spatial structures. AdaFre employs a multi-frequency routing mechanism to dynamically select and aggregate the most informative temporal frequency components, while associating each with its corresponding spatial representation derived from frequency-aware embeddings. Spatiotemporal backbones are then used to model each path independently before final aggregation. Extensive experiments on several real-world datasets demonstrate that AdaFre significantly outperforms state-of-the-art baselines.

Dynamic recommendation systems aim to provide personalized suggestions by modeling temporal user-item interactions across time-series behavioral data. Recent studies have leveraged pre-trained dynamic graph neural networks (GNNs) to learn user-item representations over temporal snapshot graphs. However, fine-tuning GNNs on these graphs often results in generalization issues due to temporal discrepancies between pre-training and fine-tuning stages, limiting the model’s ability to capture evolving user preferences. To address this, we propose TarDGR, a task-aware retrieval-augmented framework designed to enhance generalization capability by incorporating task-aware model and retrieval-augmentation. Specifically, TarDGR introduces a Task-Aware Evaluation Mechanism to identify semantically relevant historical subgraphs, enabling the construction of task-specific datasets without manual labeling. It also presents a Graph Transformer-based Task-Aware Model that integrates semantic and structural encodings to assess subgraph relevance. During inference, TarDGR retrieves and fuses task-aware subgraphs with the query subgraph, enriching its representation and mitigating temporal generalization issues. Experiments on multiple large-scale dynamic graph datasets demonstrate that TarDGR consistently outperforms state-of-the-art methods, with extensive empirical evidence underscoring its superior accuracy and generalization capabilities.

Hongxin Ding, Baixiang Huang, Yue Fang, Weibin Liao, Xinke Jiang, Jinyang Zhang, Yinghao Zhu, Zheng Li, Liantao Ma, Junfeng Zhao, Yasha Wang. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Tao Feng, Xinke Jiang, Xinyan Hu, Yonggang Zhang, Zhen Tao, Wentao Zhang, Boyang Liu, Wenhao Jiang, Chao Wu. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Zhibang Yang, Xinke Jiang, Rihong Qiu, Ruiqing Li, Yihang Zhang, Yue Fang, Yongxin Xu, Hongxin Ding, Xu Chu, Junfeng Zhao, Yasha Wang. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

By integrating external knowledge, Retrieval-Augmented Generation (RAG) has become an effective strategy for mitigating the hallucination problems that large language models (LLMs) encounter when dealing with knowledge-intensive tasks. However, in the process of integrating external non-parametric supporting evidence with internal parametric knowledge, inevitable knowledge conflicts may arise, leading to confusion in the model's responses. To enhance the knowledge selection of LLMs in various contexts, some research has focused on refining their behavior patterns through instruction-tuning. Nonetheless, due to the absence of explicit negative signals and comparative objectives, models fine-tuned in this manner may still exhibit undesirable behaviors such as contextual ignorance and contextual overinclusion. To this end, we propose a Knowledge-aware Preference Optimization strategy, dubbed KnowPO, aimed at achieving adaptive knowledge selection based on contextual relevance in real retrieval scenarios. Concretely, we proposed a general paradigm for constructing knowledge conflict datasets, which comprehensively cover various error types and learn how to avoid these negative signals through preference optimization methods. Simultaneously, we proposed a rewriting strategy and data ratio optimization strategy to address preference imbalances. Experimental results show that KnowPO outperforms previous methods for handling knowledge conflicts by over 37%, while also exhibiting robust generalization across various out-of-distribution datasets.

As a typical problem of Spatiotemporal Resource Management, Time Series Supplier Allocation (TSSA) poses a complex NP-hard challenge, aimed at refining future order dispatching strategies to satisfy the trade-off between demands and maximum supply. The Black-Litterman (BL) model, which comes from financial portfolio management, offers a new perspective for the TSSA by balancing expected returns against insufficient supply risks. However, the BL model is not only constrained by manually constructed perspective matrices and spatio-temporal market dynamics but also restricted by the absence of supervisory signals and unreliable supplier data. To solve these limitations, we introduce the pioneering Deep Black-Litterman Model for TSSA, which innovatively adapts the BL model from financial domain to supply chain context. Specifically, DBLM leverages Spatio-Temporal Graph Neural Networks (STGNNs) to capture spatio-temporal dependencies for automatically generating future perspective matrices. Moreover, a novel Spearman rank correlation is designed as our DBLM supervise signal to navigate complex risks and interactions of the supplier. Finally, DBLM further uses a masking mechanism to counteract the bias of unreliable data, thus improving precision and reliability. Extensive experiments on two datasets demonstrate significant improvements of DBLM on TSSA.

Exploring the correlations between medical features is essential for extracting patient health patterns from electronic health records (EHR) data, and strengthening medical predictions and decision-making. To constrain the hypothesis space of pure data-driven deep learning in the context of limited annotated data, a common trend is to incorporate external knowledge, especially knowledge priors related to personalized health contexts, to optimize model training. However, most existing methods lack flexibility and are constrained by the uncertainties brought about by fixed feature correlation priors. In addition, in utilizing knowledge, these methods overlook the knowledge informative for personalized healthcare. To this end, we propose DearLLM, a novel and effective framework that leverages feature correlations deduced by large language models (LLMs) to enhance personalized healthcare. Concretely, DearLLM captures and learns quantitative correlations between medical features by calculating the conditional perplexity of LLMs’ deduction based on personalized patient backgrounds. Then, DearLLM enhances healthcare predictions by emphasizing knowledge that carries unique patient information through a feature-frequency-aware graph pooling method. Extensive experiments on two real-world benchmark datasets show significant performance gains brought by DearLLM. Furthermore, the discovered findings align well with medical literature, offering meaningful clinical interpretations.

Xinke Jiang, Ruizhe Zhang, Yongxin Xu, Rihong Qiu, Yue Fang, Zhiyuan Wang, Jinyi Tang, Hongxin Ding, Xu Chu, Junfeng Zhao, Yasha Wang. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Xinke Jiang, Yue Fang, Rihong Qiu, Haoyu Zhang, Yongxin Xu, Hao Chen, Wentao Zhang, Ruizhe Zhang, Yuchen Fang, Xinyu Ma, Xu Chu, Junfeng Zhao, Yasha Wang. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Yongxin Xu, Ruizhe Zhang, Xinke Jiang, Yujie Feng, Yuzhen Xiao, Xinyu Ma, Runchuan Zhu, Xu Chu, Junfeng Zhao, Yasha Wang. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Hongxin Ding, Yue Fang, Runchuan Zhu, Xinke Jiang, Jinyang Zhang, Yongxin Xu, Weibin Liao, Xu Chu, Junfeng Zhao, Yasha Wang. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.

Graph Neural Networks (GNNs) have emerged as a fundamental tool for modeling complex graph structures across diverse applications. However, directly applying pretrained GNNs to varied downstream tasks without fine-tuning-based continual learning remains challenging, as this approach incurs high computational costs and hinders the development of Large Graph Models (LGMs). In this paper, we investigate an efficient and generalizable dataset distillation framework for Graph Continual Learning (GCL) across multiple downstream tasks, implemented through a novel Lightweight Graph Neural Tangent Kernel (LIGHTGNTK). Specifically, LIGHTGNTK employs a low-rank approximation of the Laplacian matrix via Bernoulli sampling and linear association within the GNTK. This design enables efficient capture of both structural and feature relationships while supporting gradient-based dataset distillation. Additionally, LIGHTGNTK incorporates a unified subgraph anchoring strategy, allowing it to handle graph-level, node-level, and edge-level tasks under diverse input structures. Comprehensive experiments on several datasets show that LIGHTGNTK achieves state-of-the-art performance in GCL scenarios, promoting the development of adaptive and scalable LGMs.

Road traffic forecasting is crucial in real-world intelligent transportation scenarios like traffic dispatching and path planning in city management and personal traveling. Spatio-temporal graph neural networks (STGNNs) stand out as the mainstream solution in this task. Nevertheless, the quadratic complexity of remarkable dynamic spatial modeling-based STGNNs has become the bottleneck over large-scale traffic data. From the spatial data management perspective, we present a novel Transformer framework called PatchSTG to efficiently and dynamically model spatial dependencies for large-scale traffic forecasting with interpretability and fidelity. Specifically, we design a novel irregular spatial patching to reduce the number of points involved in the dynamic calculation of Transformer. The irregular spatial patching first utilizes the leaf K-dimensional tree (KDTree) to recursively partition irregularly distributed traffic points into leaf nodes with a small capacity, and then merges leaf nodes belonging to the same subtree into occupancy-equaled and non-overlapped patches through padding and backtracking. Based on the patched data, depth and breadth attention are used interchangeably in the encoder to dynamically learn local and global spatial knowledge from points in a patch and points with the same index of patches. Experimental results on four real world large-scale traffic datasets show that our PatchSTG achieves train speed and memory utilization improvements up to 10x and 4x with the state-of-the-art performance.

Measuring parameter importance is crucial for understanding and optimizing large language models (LLMs). Existing work predominantly focuses on pruning or probing at neuron/feature levels without fully considering the cooperative behaviors of model parameters. In this paper, we introduce a novel approach--Model Shapley to quantify parameter importance based on the Shapley value, a principled method from cooperative game theory that captures both individual and synergistic contributions among parameters, via only one gradient backpropagation. We derive a scalable second-order approximation to compute Shapley values at the parameter level, leveraging blockwise Fisher information for tractability in large-scale settings. Our method enables fine-grained differentiation of parameter importance, facilitating targeted knowledge injection and model compression. Through mini-batch Monte Carlo updates and efficient approximation of the Hessian structure, we achieve robust Shapley-based attribution with only modest computational overhead. Experimental results indicate that this cooperative game perspective enhances interpretability, guides more effective parameter-specific fine-tuning and model compressing, and paves the way for continuous model improvement in various downstream tasks.

Spatio-Temporal Graph Neural Networks (STGNNs) have emerged as a powerful tool for modeling dynamic graph-structured data across diverse domains. However, they often fail to generalize in Spatio-Temporal Out-of-Distribution (STOOD) scenarios, where both temporal dynamics and spatial structures evolve beyond the training distribution. To address this problem, we propose an innovative Spatio-Temporal Retrieval-Augmented Pattern Learning framework, STRAP, which enhances model generalization by integrating retrieval-augmented learning into the STGNN continue learning pipeline. The core of STRAP is a compact and expressive pattern library that stores representative spatio-temporal patterns enriched with historical, structural, and semantic information, which is obtained and optimized during the training phase. During inference, STRAP retrieves relevant patterns from this library based on similarity to the current input and injects them into the model via a plug-and-play prompting mechanism. This not only strengthens spatio-temporal representations but also mitigates catastrophic forgetting. Moreover, STRAP introduces a knowledge-balancing objective to harmonize new information with retrieved knowledge. Extensive experiments across multiple real-world streaming graph datasets show that STRAP consistently outperforms state-of-the-art STGNN baselines on STOOD tasks, demonstrating its robustness, adaptability, and strong generalization capability without task-specific fine-tuning.

Multi-domain aspect-based sentiment analysis (ABSA) seeks to capture fine-grained sentiment across diverse domains. While existing research narrowly focuses on single-domain applications constrained by methodological limitations and data scarcity, the reality is that sentiment naturally traverses multiple domains. Although large language models (LLMs) offer a promising solution for ABSA, it is difficult to integrate effectively with established techniques, including graph-based models and linguistics, because modifying their internal architecture is not easy. To alleviate this problem, we propose a novel framework, Feature-aware In-context Learning for Multi-domain ABSA (FaiMA). The core insight of FaiMA is to utilize in-context learning (ICL) as a feature-aware mechanism that facilitates adaptive learning in multi-domain ABSA tasks. Specifically, we employ a multi-head graph attention network as a text encoder optimized by heuristic rules for linguistic, domain, and sentiment features. Through contrastive learning, we optimize sentence representations by focusing on these diverse features. Additionally, we construct an efficient indexing mechanism, allowing FaiMA to stably retrieve highly relevant examples across multiple dimensions for any given input. To evaluate the efficacy of FaiMA, we build the first multi-domain ABSA benchmark dataset. Extensive experimental results demonstrate that FaiMA achieves significant performance improvements in multiple domains compared to baselines, increasing F1 by 2.07% on average. Source code and data sets are available at https://github.com/SupritYoung/FaiMA.

With the widespread adoption of electronic health records (EHR) data, deep learning techniques have been broadly utilized for various health prediction tasks. Nevertheless, the labeled data scarcity issue restricts the prediction power of these deep models. To enhance the generalization capability of deep learning models when faced with such situations, a common trend is to train generative adversarial networks (GANs) or diffusion models for data augmentation. However, due to limitations in sample size and potential label imbalance issues, these methods are prone to mode collapse problems. This results in the generation of new samples that fail to preserve the subtype structure within EHR data, thereby limiting their practicality in health prediction tasks that generally require detailed patient phenotyping. Aiming at the above problems, we propose a Prototype-based Mixup method, dubbed ProtoMix, which combines prior knowledge of intrinsic data features from subtype centroids (i.e., prototypes) to guide the synthesis of new samples. Specifically, ProtoMix employs a prototype-guided mixup training task to shift the decision boundary away from the subtypes. Then, ProtoMix optimizes the sampling weights in different areas of the data manifold via a prototype-guided mixup sampling strategy. Throughout the training process, ProtoMix dynamically expands the training distribution using an adaptive mixing coefficient computation method. Experimental evaluations on three real-world datasets demonstrate the efficacy of ProtoMix.

Graph Neural Networks (GNNs) have become essential in interpreting relational data across various domains, yet, they often struggle to generalize to unseen graph data that differs markedly from training instances. In this paper, we introduce a novel framework called General Retrieval-Augmented Graph Learning (RAGraph), which brings external graph data into the general graph foundation model to improve model generalization on unseen scenarios. On the top of our framework is a toy graph vector library that we established, which captures key attributes, such as features and task-specific label information. During inference, the RAGraph adeptly retrieves similar toy graphs based on key similarities in downstream tasks, integrating the retrieved data to enrich the learning context via the message-passing prompting mechanism. Our extensive experimental evaluations demonstrate that RAGraph significantly outperforms state-of-the-art graph learning methods in multiple tasks such as node classification, link prediction, and graph classification across both dynamic and static datasets. Furthermore, extensive testing confirms that RAGraph consistently maintains high performance without the need for task-specific fine-tuning, highlighting its adaptability, robustness, and broad applicability.

Graph Neural Networks (GNNs) conventionally operate under the assumption that node attributes are entirely observable. Their performance notably deteriorates when confronted with incomplete graphs due to the inherent message-passing mechanisms. Current solutions either employ classic imputation techniques or adapt GNNs to tolerate missed attributes. However, their ability to generalize is impeded especially when dealing with high rates of missing attributes. To address this, we harness the representations of the essential views on graphs, attributes and structures, into a common shared latent space, ensuring robust tolerance even at high missing rates. Our proposed neural model, named ASD-VAE, parameterizes such space via a coupled-and-decoupled learning procedure, reminiscent of brain cognitive processes and multimodal fusion. Initially, ASD-VAE separately encodes attributes and structures, generating representations for each view. A shared latent space is then learned by maximizing the likelihood of the joint distribution of different view representations through coupling. Then, the shared latent space is decoupled into separate views, and the reconstruction loss of each view is calculated. Finally, the missing values of attributes are imputed from this learned latent space. In this way, the model offers enhanced resilience against skewed and biased distributions typified by missing information and subsequently brings benefits to downstream graph machine-learning tasks. Extensive experiments conducted on four typical real-world incomplete graph datasets demonstrate the superior performance of ASD-VAE against the state-of-the-art

Understanding Origin-Destination (O-D) travel demand is crucial for transportation management. However, traditional spatial-temporal deep learning models grapple with addressing the sparse and long-tail characteristics in high-resolution O-D matrices and quantifying prediction uncertainty. This dilemma arises from the numerous zeros and over-dispersed demand patterns within these matrices, which challenge the Gaussian assumption inherent to deterministic deep learning models. To address these challenges, we propose a novel approach: the Spatial-Temporal Tweedie Graph Neural Network (STTD). The STTD introduces the Tweedie distribution as a compelling alternative to the traditional 'zero-inflated' model and leverages spatial and temporal embeddings to parameterize travel demand distributions. Our evaluations using real-world datasets highlight STTD's superiority in providing accurate predictions and precise confidence intervals, particularly in high-resolution scenarios. GitHub code is available online(https://github.com/STTDAnonymous/STTD).

Modeling complex spatial and temporal dependencies are indispensable for location-bound time series learning. Existing methods, typically relying on graph neural networks (GNNs) and temporal learning modules based on recurrent neural networks, have achieved significant performance improvements. However, their representation capabilities and prediction results are limited when pre-defined graphs are unavailable. Unlike spatio-temporal GNNs focusing on designing complex architectures, we propose a novel adaptive graph construction strategy: Self-Paced Graph Contrastive Learning (SPGCL). It learns informative relations by maximizing the distinguishing margin between positive and negative neighbors and generates an optimal graph with a self-paced strategy. Specifically, the existing neighborhoods iteratively absorb more reliable nodes with the highest affinity scores as new neighbors to generate the next-round neighborhoods, and augmentations are applied to improve the transferability and robustness. As the adaptively self-paced graph approaches the optimized graph for prediction, the mutual information between nodes and the corresponding neighbors is maximized. Our work provides a new perspective of addressing spatio-temporal learning problems beyond information aggregation in Euclidean space and can be generalized to different tasks. Extensive experiments conducted on two typical spatio-temporal learning tasks (traffic forecasting and land displacement prediction) demonstrate the superior performance of SPGCL against the state-of-the-art.