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.

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.

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.

Yujie Feng, Xujia Wang, Zexin Lu, Shenghong Fu, Guangyuan Shi, Yongxin Xu, Yasha Wang, Philip S. Yu, Xu Chu, Xiao-Ming Wu. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

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.

Yujie Feng, Li-Ming Zhan, Zexin Lu, Yongxin Xu, Xu Chu, Yasha Wang, Jiannong Cao, Philip S. Yu, Xiao-Ming Wu. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.

We study online task scheduling problems where tasks arrive sequentially and are processed by the platform or server. The service processes for tasks are multi-stage and are modeled as episodic Markov Decision Processes (MDPs). While processing a task, the system acquires rewards by consuming resources. The goal of the platform is to maximize the reward-to-cost ratio over a sequence of K tasks. Online scheduling with multi-stage tasks faces two major challenges: intra-dependence among the different stages within a task and inter-dependence among different tasks. These challenges are further exacerbated by the unknown rewards, costs, and task arrival distribution. To address these challenges, we propose the Robbins-Monro-based Value Iteration for Ratio Maximization (RM^2VI) algorithm. Specifically,RM^2VI addresses ``intra-dependence'' through optimistic value iteration and handles ``inter-dependence'' using the Robbins-Monro method. The algorithm has a greedy structure and achieves a sub-linear regret of O(K^(3/4)), establishing the no-regret property (per-task). We test RM^2VI in two synthetic experiments of sale promotion in E-commerce and machine learning job training in cloud computing. The results show RM^2VI achieves the best reward-to-cost ratio compared with the baselines.

A practical dialogue system requires the capacity for ongoing skill acquisition and adaptability to new tasks while preserving prior knowledge.However, current methods for Continual Dialogue State Tracking (DST), a crucial function of dialogue systems, struggle with the catastrophic forgetting issue and knowledge transfer between tasks.We present TaSL, a novel framework for task skill localization and consolidation that enables effective knowledge transfer without relying on memory replay.TaSL uses a novel group-wise technique to pinpoint task-specific and task-shared areas.Additionally, a fine-grained skill consolidation strategy protects task-specific knowledge from being forgotten while updating shared knowledge for bi-directional knowledge transfer.As a result, TaSL strikes a balance between preserving previous knowledge and excelling at new tasks.Comprehensive experiments on various backbones highlight the significant performance improvements of TaSL over existing state-of-the-art methods.The source code 1 is provided for reproducibility.

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.

While recent developments of deep learning models have led to record-breaking achievements in many areas, the lack of sufficient interpretation remains a problem for many specific applications, such as the diagnosis prediction task in healthcare. The previous knowledge graph(KG) enhanced approaches mainly focus on learning clinically meaningful representations, the importance of medical concepts, and even the knowledge paths from inputs to labels. However, it is infeasible to interpret the diagnosis prediction, which needs to consider different medical concepts, various medical relationships, and the time-effectiveness of knowledge triples in different patient contexts. More importantly, the retrospective and prospective interpretations of disease processes are valuable to clinicians for the patients' confounding diseases. We propose KerPrint, a novel KG enhanced approach for retrospective and prospective interpretations to tackle these problems. Specifically, we propose a time-aware KG attention method to solve the problem of knowledge decay over time for trustworthy retrospective interpretation. We also propose a novel element-wise attention method to select candidate global knowledge using comprehensive representations from the local KG for prospective interpretation. We validate the effectiveness of our KerPrint through an extensive experimental study on a real-world dataset and a public dataset. The results show that our proposed approach not only achieves significant improvement over knowledge-enhanced methods but also gives the interpretability of diagnosis prediction in both retrospective and prospective views.

Due to the insufficiency of electronic health records (EHR) data utilized in practical diagnosis prediction scenarios, most works are devoted to learning powerful patient representations either from structured EHR data (e.g., temporal medical events, lab test results, etc.) or unstructured data (e.g., clinical notes, etc.). However, synthesizing rich information from both of them still needs to be explored. Firstly, the heterogeneous semantic biases across them heavily hinder the synthesis of representation spaces, which is critical for diagnosis prediction. Secondly, the intermingled quality of partial clinical notes leads to inadequate representations of to-be-predicted patients. Thirdly, typical attention mechanisms mainly focus on aggregating information from similar patients, ignoring important auxiliary information from others. To tackle these challenges, we propose a novel visit sequences-clinical notes joint learning approach, dubbed VecoCare. It performs a Gromov-Wasserstein Distance (GWD)-based contrastive learning task and an adaptive masked language model task in a sequential pre-training manner to reduce heterogeneous semantic biases. After pre-training, VecoCare further aggregates information from both similar and dissimilar patients through a dual-channel retrieval mechanism. We conduct diagnosis prediction experiments on two real-world datasets, which indicates that VecoCare outperforms state-of-the-art approaches. Moreover, the findings discovered by VecoCare are consistent with the medical researches.

Deep learning techniques are capable of capturing complex input-output relationships, and have been widely applied to the diagnosis prediction task based on web-based patient electronic health records (EHR) data. To improve the prediction and interpretability of pure data-driven deep learning with only a limited amount of labeled data, a pervasive trend is to assist the model training with knowledge priors from online medical knowledge graphs. However, they marginally investigated the label imbalance and the task-irrelevant noise in the external knowledge graph. The imbalanced label distribution would bias the learning and knowledge extraction towards the majority categories. The task-irrelevant noise introduces extra uncertainty to the model performance. To this end, aiming at by-passing the bias-variance trade-off dilemma, we introduce a new sequential learning framework, dubbed SeqCare, for diagnosis prediction with online medical knowledge graphs. Concretely, in the first step, SeqCare learns a bias-reduced space through a self-supervised graph contrastive learning task. Secondly, SeqCare reduces the learning uncertainty by refining the supervision signal and the graph structure of the knowledge graph simultaneously. Lastly, SeqCare trains the model in the bias-variance reduced space with a self-distillation to further filter out irrelevant information in the data. Experimental evaluations on two real-world datasets show that SeqCare outperforms state-of-the-art approaches. Case studies exemplify the interpretability of SeqCare. Moreover, the medical findings discovered by SeqCare are consistent with experts and medical literature.

Individuals and organizations are using databases to store personal information at an unprecedented rate. This creates a quandary for data providers. They are responsible for protecting the privacy of individuals described in their database. On the other hand, data providers are sometimes required to provide statistics about their data instead of sharing it wholesale with strong assurances that these answers are correct and complete such as in regulatory filings for the US SEC and other goverment organizations. We introduce a system, ZKSQL , that provides authenticated answers to ad-hoc SQL queries with zero-knowledge proofs. Its proofs show that the answers are correct and sound with respect to the database's contents and they do not divulge any information about its input records. This system constructs proofs over the steps in a query's evaluation and it accelerates this process with authenticated set operations. We validate the efficiency of this approach over a suite of TPC-H queries and our results show that ZKSQL achieves two orders of magnitude speedup over the baseline.