论文检索

Zixiang Wang, Yinghao Zhu, Huiya Zhao, Xiaochen Zheng, Dehao Sui, Tianlong Wang, Wen Tang 0001, Yasha Wang, Ewen M. Harrison, Chengwei Pan 等

We introduce ColaCare, a framework that enhances Electronic Health Record (EHR) modeling through multi-agent collaboration driven by Large Language Models (LLMs). Our approach seamlessly integrates domain-specific expert models with LLMs to bridge the gap between structured EHR data and text-based reasoning. Inspired by the Multidisciplinary Team (MDT) approach used in clinical settings, ColaCare employs two types of agents: DoctorAgents and a MetaAgent, which collaboratively analyze patient data. Expert models process and generate predictions from numerical EHR data, while LLM agents produce reasoning references and decision-making reports within the MDT-driven collaborative consultation framework. The MetaAgent orchestrates the discussion, facilitating consultations and evidence-based debates among DoctorAgents, simulating diverse expertise in clinical decision-making. We additionally incorporate the Merck Manual of Diagnosis and Therapy (MSD) medical guideline within a retrieval-augmented generation (RAG) module for medical evidence support, addressing the challenge of knowledge currency. Extensive experiments conducted on three EHR datasets demonstrate ColaCare's superior performance in clinical mortality outcome and readmission prediction tasks, underscoring its potential to revolutionize clinical decision support systems and advance personalized precision medicine. All code, case studies and a questionnaire are available at the project website: https://colacare.netlify.app.

Yayao Hong, Liyue Chen, Leye Wang, Xiuhuai Xie, Guofeng Luo, Cheng Wang 0003, Longbiao Chen

Ubiquitous sensors and mobile devices have spurred the growth of Web-of-Things (WoT) services in smart cities, making accurate spatio-temporal traffic predictions increasingly crucial. Leveraging advances in deep learning, recent Spatio-Temporal Graph Neural Networks (STGNNs) have achieved remarkable results. However, these methods address scenario-specific spatio-temporal heterogeneity by designing model architectures, often overlooking the importance of selecting optimal spatio-temporal knowledge (i.e., model inputs). In this paper, we propose an automated framework for spatio-temporal knowledge optimization to address this challenge. Our framework seamlessly integrates with downstream models, enhancing their performance across various prediction tasks. Specifically, we design a knowledge search space composed of parameters that represent scenario-specific spatio-temporal correlations within data. Additionally, we employ a bandit-based multi-fidelity algorithm for knowledge optimization to solve the constraint of limited resource. Furthermore, we adopt a meta-learner to extract transferable meta-knowledge about optimal knowledge, facilitating efficient exploration of the search space. Extensive experiments on five widely used real-world datasets demonstrate the effectiveness of our proposed framework. To the best of our knowledge, we are the first to automatically optimize spatio-temporal knowledge for spatio-temporal traffic prediction.

Zhengliang Shi, Shen Gao, Lingyong Yan, Yue Feng 0002, Xiuyi Chen, Zhumin Chen, Dawei Yin 0001, Suzan Verberne, Zhaochun Ren

Augmenting large language models (LLMs) with external tools has emerged as a promising approach to extend their utility, enabling them to solve practical tasks. Previous methods manually parse tool documentation and create in-context demonstrations, transforming tools into structured formats for LLMs to use in their step-by-step reasoning. However, this manual process requires domain expertise and struggles to scale to large toolsets. Additionally, these methods rely heavily on ad-hoc inference techniques or special tokens to integrate free-form LLM generation with tool-calling actions, limiting the LLM's flexibility in handling diverse tool specifications and integrating multiple tools. In this work, we propose AutoTools, a framework that enables LLMs to automate the tool-use workflow. Specifically, the LLM automatically transforms tool documentation into callable functions, verifying syntax and runtime correctness. Then, the LLM integrates these functions into executable programs to solve practical tasks, flexibly grounding tool-use actions into its reasoning processes. Extensive experiments on existing and newly collected, more challenging benchmarks illustrate the superiority of our framework. Inspired by these promising results, we further investigate how to improve the expertise of LLMs, especially open-source LLMs with fewer parameters, within AutoTools. Thus, we propose the AutoTools-Learning approach, training the LLMs with three learning tasks on 34k instances of high-quality synthetic data, including documentation understanding, relevance learning, and function programming. Fine-grained results validate the effectiveness of our overall training approach and each individual task. Our methods are an important step towards the use of LLMs for solving real-world tasks with external tools.

Sujia Huang, Yueyang Pi, Tong Zhang 0021, Wenzhe Liu, Zhen Cui 0001

Graph Neural Networks (GNNs) have expressed remarkable capability in processing graph-structured data. Recent studies have found that most GNNs rely on the homophily assumption of graphs, leading to unsatisfactory performance on heterophilous graphs. While certain methods have been developed to address heterophilous links, they lack more precise estimation of high-order relationships between nodes. This could result in the aggregation of excessive interference information during message propagation, thus degrading the representation ability of learned features. In this work, we propose a Disparity-induced Structural Refinement (DSR) method that enables adaptive and selective message propagation in GNN, to enhance representation learning in heterophilous graphs. We theoretically analyze the necessity of structural refinement during message passing grounded in the derivation of error bound for node classification. To this end, we design a disparity score that combines both features and structural information at the node level, reflecting the connectivity degree of hopping neighbor nodes. Based on the disparity score, we can adjust the aggregation of neighbor nodes, thereby mitigating the impact of irrelevant information during message passing. Experimental results demonstrate that our method achieves competitive performance, mostly outperforming advanced methods on both homophilous and heterophilous datasets.

Yi Zhang 0103, Yiwen Zhang 0001

The core of the general recommender systems lies in learning high-quality embedding representations of users and items to investigate their positional relations in the feature space. Unfortunately, data sparsity caused by difficult-to-access interaction data severely limits the effectiveness of recommender systems. Faced with such a dilemma, various types of self-supervised learning methods have been introduced into recommender systems in an attempt to alleviate the data sparsity through distribution modeling or data augmentation. However, most data augmentation relies on elaborate manual design, which is not only not universal, but the bloated and redundant augmentation process may significantly slow down model training progress. To tackle these limitations, we propose a novel Dual Mixing-based Recommendation Framework (MixRec) to empower data augmentation as we wish. Specifically, we propose individual mixing and collective mixing, respectively. The former aims to provide a new positive sample that is unique to the target (user or item) and to make the pair-wise recommendation loss benefit from it, while the latter aims to portray a new sample that contains group properties in a batch. The two mentioned mixing mechanisms allow for data augmentation with only one parameter that does not need to be set multiple times and can be done in linear time complexity. Besides, we propose the dual-mixing contrastive learning to maximize the utilization of these new-constructed samples to enhance the consistency between pairs of positive samples. Experimental results on four real-world datasets demonstrate the advantages of MixRec in terms of effectiveness, simplicity, efficiency, and scalability.

Yun Zhu 0007, Haizhou Shi, Xiaotang Wang, Yongchao Liu 0004, Yaoke Wang, Boci Peng, Chuntao Hong, Siliang Tang

Recently, research on Text-Attributed Graphs (TAGs) has gained significant attention due to the prevalence of free-text node features in real-world applications and the advancements in Large Language Models (LLMs) that bolster TAG methodologies. However, current TAG approaches face two primary challenges: (i) Heavy reliance on label information and (ii) Limited cross-domain zero/few-shot transferability. These issues constrain the scaling of both data and model size, owing to high labor costs and scaling laws, complicating the development of graph foundation models with strong transferability. In this work, we propose the GraphCLIP framework to address these challenges by learning graph foundation models with strong cross-domain zero/few-shot transferability through a self-supervised contrastive graph-summary pretraining method. Specifically, we generate and curate large-scale graph-summary pair data with the assistance of LLMs, and introduce a novel graph-summary pretraining method, combined with invariant learning, to enhance graph foundation models with strong cross-domain zero-shot transferability. For few-shot learning, we propose a novel graph prompt tuning technique aligned with our pretraining objective to mitigate catastrophic forgetting and minimize learning costs. Extensive experiments show the superiority of GraphCLIP in both zero-shot and few-shot settings, while evaluations across various downstream tasks confirm the versatility of GraphCLIP. Our code is available at: https://github.com/ZhuYun97/GraphCLIP.

Weiheng Wu, Wei Qiao 0005, Wenhao Yan, Bo Jiang 0013, Yuling Liu, Baoxu Liu, Zhigang Lu 0002, Junrong Liu

Advanced Persistent Threats (APTs) are continuously evolving, leveraging their stealthiness and persistence to put increasing pressure on current provenance-based Intrusion Detection Systems (IDS). This evolution exposes several critical issues: (1) The dense interaction between malicious and benign nodes within provenance graphs introduces neighbor noise, hindering effective detection; (2) The complex prediction mechanisms of existing APTs detection models lead to the insufficient utilization of prior knowledge embedded in the data; (3) The high computational cost makes detection impractical. To address these challenges, we propose Vodka, a lightweight threat detection system built on a knowledge distillation framework, capable of node-level detection within audit log provenance graphs. Specifically, Vodka applies graph Laplacian regularization to reduce neighbor noise, obtaining smoothed and denoised graph signals. Subsequently, Vodka employs a teacher model based on GNNs to extract knowledge, which is then distilled into a lightweight student model. The student model is designed as a combination of a feature transformation module and a personalized PageRank random walk label propagation module, with the former capturing feature knowledge and the latter learning label and structural knowledge. After distillation, the student model benefits from the knowledge of the teacher model to perform precise threat detection. Finally, Vodka reconstructs attack paths from anomalous nodes, providing insight into the attackers' strategies. We evaluate Vodka through extensive experiments on three public datasets and compare its performance against several state-of-the-art IDS solutions. The results demonstrate that Vodka achieves outstanding detection accuracy across all scenarios.

Xuanwen Huang, Wei Chow, Yize Zhu, Yang Wang, Ziwei Chai, Chunping Wang 0001, Lei Chen 0082, Yang Yang 0009

This paper proposes CrossLink, a novel framework for cross-domain link prediction. CrossLink learns the evolution pattern of a specific downstream graph and subsequently makes pattern-specific link predictions. It employs a technique called conditioned link generation, which integrates both evolution and structure modeling to perform evolution-specific link prediction. This conditioned link generation is carried out by a transformer-decoder architecture, enabling efficient parallel training and inference. CrossLink is trained on extensive dynamic graphs across diverse domains, encompassing 6 million dynamic edges. Extensive experiments on eight untrained graphs demonstrate that CrossLink achieves state-of-the-art performance in cross-domain link prediction. Compared to advanced baselines under the same settings, CrossLink shows an average improvement of 11.40% in Average Precision across eight graphs. Impressively, it surpasses the fully supervised performance of 8 advanced baselines on 6 untrained graphs. Project Page is https://zjunet.github.io/CrossLink/

Shuzheng Wang, Yue Huang, Wenqin Zhang, Yuming Huang 0002, Xuechao Wang, Jing Tang 0004

Ethereum, as a representative of Web3, adopts a novel framework called Proposer Builder Separation (PBS) to prevent the centralization of block profits in the hands of institutional Ethereum stakers. Introducing builders to generate blocks based on public transactions, PBS aims to ensure that block profits are distributed among all stakers. Through the auction among builders, only one will win the block in each slot. Ideally, the equilibrium strategy of builders under public information would lead them to bid all block profits. However, builders are now capable of extracting profits from private order flows. In this paper, we explore the effect of PBS with private order flows. Specifically, we propose the asymmetry auction model of MEV-Boost auction. Moreover, we conduct empirical study on Ethereum blocks from January 2023 to May 2024. Our analysis indicates that private order flows contribute to 54.59% of the block value, indicating that different builders will build blocks with different valuations. Interestingly, we find that builders with more private order flows (i.e., higher block valuations) are more likely to win the block, while retain larger proportion of profits. In return, such builders will further attract more private order flows, resulting in a monopolistic market gradually. Our findings reveal that PBS in current stage is unable to balance the profit distribution, which just transits the centralization of block profits from institutional stakers to the monopolistic builder.

Hong Liu, Zhe Wang 0001, Kewen Wang 0001, Xiaowang Zhang, Zhiyong Feng 0002

Logical rules have been widely used for expressing schema knowledge in various practical applications. It is infeasible to handcraft rules from large knowledge graphs (KGs) and thus many methods have been proposed for learning rules automatically from KGs. However, it is largely ignored how to extract rules in a (target) KG from rules that already exist in some other (source) KGs. In this paper, we propose a framework for KG rule learning based on transfer learning. A major challenge for establishing such a framework is that a suitable alignment mechanism is required for mapping certain subgraph structures between predicates in the source KG and the target KG. Hence, our framework provides a new method for predicate mapping based on graph-structural similarity. The proposed framework can be used as a standalone rule learner but more importantly, it paves a new way for enhancing the state-of-the-art rule learners for large KGs. Extensive experiments are conducted to evaluate the new approach to rule learning, which shows that rules in smaller KGs can be effectively transferred to a large KG.

Sijie Mai, Ying Zeng, Haifeng Hu 0001

Previous studies on multimodal affective computing primarily focus on approximating predictions to annotated labels, often neglecting the ordinal nature of affective states. In this paper, we address this issue by exploring ordinal learning, and a Multimodal Ordinal Affective Computing (MOAC) framework is designed to enhance the understanding of the nature of affective concepts. Specifically, we propose coarse-grained label-level ordinal learning that prompts the model to learn to compare in the label space, encouraging higher predictive values for samples annotated with larger labels over those with smaller labels. Moreover, a regularization loss is proposed to prevent the output distributions from deviating significantly from the annotated label distributions. Fine-grained feature-level ordinal learning is then performed via the feature difference operation and the neutral embedding. The former compares samples in the feature space, calculating the difference between features of different samples to generate 'new' features for a more robust training. The latter seeks to reduce the difficulty of prediction by estimating the difference between the target multimodal representations and a neutral reference. We first demonstrate MOAC in multimodal sentiment analysis, which is a regression task that aligns well with the function of ordinal learning. Then we extend MOAC to classification tasks including multimodal humor detection and sarcasm detection to evaluate its generalizability. Experiments suggest that MOAC outperforms state-of-the-art methods.

Jianling Gao, Chongyang Tao, Zhenchao Sun, Xiya Jiang, Shuai Ma 0001

Semi-supervised anomaly detection (AD) has garnered growing attention due to its ability to effectively leverage limited labeled data to identify anomalies. However, current methods often impose artificial constraints on the proportion of unlabeled anomalies in the training set, thereby impeding the effective training of models for anomaly detection in real-world scenarios where several anomalies may be present in the unlabeled dataset. Additionally, existing methods often struggle to effectively exploit and model the complex relationships between data instances, which is critical for learning more discriminative features and accurate distance measures. Distance-based methods, in particular, typically rely on Euclidean distance metric, which lacks the flexibility to capture complex correlations across different data dimensions. To address the above challenges, we propose CAD, a denoising-aware Contrastive distance learning framework for semi-supervised AD. It introduces a contrastive training objective to facilitate the learning of distinctive representations by contrasting the average distance between anomalies and unlabeled samples. To fully exploit the information from the unlabeled data meanwhile mitigate the effects of noise, we incorporate a two-stage anomaly denoising and expansion strategy to refine the dataset by identifying high-confidence samples from the unlabeled set. Furthermore, we employ a parameterized bilinear tensor distance layer to learn a customized distance metric, enabling the model to capture intricate relationships among data points. Extensive experiments on 10 real-world datasets demonstrate that CAD significantly outperforms existing semi-supervised AD models. Code available at https://github.com/CADrepo/CAD.

Jiayi Luo, Qingyun Sun, Haonan Yuan, Xingcheng Fu, Jianxin Li 0002

Adversarial evasion attacks pose significant threats to graph learning, with lines of studies that have improved the robustness of Graph Neural Networks (GNNs). However, existing works rely on priors about clean graphs or attacking strategies, which are often heuristic and inconsistent. To achieve robust graph learning over different types of evasion attacks and diverse datasets, we investigate this problem from a prior-free structure purification perspective. Specifically, we propose a novel Diffusion-based Structure Purification framework named DiffSP, which creatively incorporates the graph diffusion model to learn intrinsic distributions of clean graphs and purify the perturbed structures by removing adversaries under the direction of the captured predictive patterns without relying on priors. DiffSP is divided into the forward diffusion process and the reverse denoising process, during which structure purification is achieved. To avoid valuable information loss during the forward process, we propose an LID-driven non-isotropic diffusion mechanism to selectively inject noise anisotropically. To promote semantic alignment between the clean graph and the purified graph generated during the reverse process, we reduce the generation uncertainty by the proposed graph transfer entropy guided denoising mechanism. Extensive experiments demonstrate the superior robustness of DiffSP against evasion attacks.

Baolei Zhang, Haoran Xin 0002, Minghong Fang, Zhuqing Liu, Biao Yi, Tong Li 0011, Zheli Liu

Large language models (LLMs) integrated with retrieval-augmented generation (RAG) systems improve accuracy by leveraging external knowledge sources. However, recent research has revealed RAG's susceptibility to poisoning attacks, where the attacker injects poisoned texts into the knowledge database, leading to attacker-desired responses. Existing defenses, which predominantly focus on inference-time mitigation, have proven insufficient against sophisticated attacks. In this paper, we introduce RAGForensics, the first traceback system for RAG, designed to identify poisoned texts within the knowledge database that are responsible for the attacks. RAGForensics operates iteratively, first retrieving a subset of texts from the database and then utilizing a specially crafted prompt to guide an LLM in detecting potential poisoning texts. Empirical evaluations across multiple datasets demonstrate the effectiveness of RAGForensics against state-of-the-art poisoning attacks. This work pioneers the traceback of poisoned texts in RAG systems, providing a practical and promising defense mechanism to enhance their security.

Shenghao Yang 0004, Weizhi Ma, Zhiqiang Guo, Min Zhang 0006, Haiyang Wu, Junjie Zhai, Chunhui Zhang, Yuekui Yang

With the development of multi-modal modeling techniques, recent sequential recommender systems enhance transferability by incorporating cross-domain universal multi-modal data, e.g., text and image. Existing methods typically adopt pairwise alignment to alleviate the gap between modalities. However, this alignment paradigm has limitations on explainability, consistency, and expansibility, resulting in suboptimal performance. This paper proposes a novel Explainable multi-modality Alignment method for transferable Rec ommender systems, i.e., EARec. Specifically, we design a two-stage framework to achieve explainable modality alignment in the source domain and recommendation based on aligned modality representations in the target domain. In the first stage, we adopt a generative task to align various modalities in parallel to a shared anchor with explainable meaning. All modalities share the same anchor to ensure consistent direction. Additionally, we treat behavior as an independent modality to integrate task-specific information into the alignment framework. In the second stage, we compose multiple item modality representation models trained in the first stage to obtain a unified model capable of understanding various modalities simultaneously, thereby providing high-quality item modality representations for recommendations in the target domain. Benefiting from the approach of parallel modality alignment followed by model composition, the framework shows flexibility in expanding new modalities. Experimental results on multiple public datasets demonstrate the superiority of EARec over baselines, and further analyses indicate the explainability and expansibility of the proposed alignment method.

Qi Yu, Zhichen Zeng 0001, Yuchen Yan, Lei Ying 0001, R. Srikant 0001, Hanghang Tong

Network alignment, which aims to find node correspondence across different networks, is the cornerstone of various downstream multi-network and Web mining tasks. Most of the embedding-based methods indirectly model cross-network node relationships by contrasting positive and negative node pairs sampled from hand-crafted strategies, which are vulnerable to graph noises and lead to potential misalignment of nodes. Another line of work based on the optimal transport (OT) theory directly models cross-network node relationships and generates noise-reduced alignments. However, OT methods heavily rely on fixed, pre-defined cost functions that prohibit end-to-end training and are hard to generalize. In this paper, we aim to unify the embedding and OT-based methods in a mutually beneficial manner and propose a joint optimal transport and embedding framework for network alignment named JOENA. For one thing (OT for embedding), through a simple yet effective transformation, the noise-reduced OT mapping serves as an adaptive sampling strategy directly modeling all cross-network node pairs for robust embedding learning. For another (embedding for OT), on top of the learned embeddings, the OT cost can be gradually trained in an end-to-end fashion, which further enhances the alignment quality. With a unified objective, the mutual benefits of both methods can be achieved by an alternating optimization schema with guaranteed convergence. Extensive experiments on real-world networks validate the effectiveness and scalability of JOENA, achieving up to 16% improvement in MRR and 20 times speedup compared with the state-of-the-art alignment methods.

Xutong Wang, Yun Feng 0003, Bingsheng Bi, Yaqin Cao, Ze Jin, Xinyu Liu 0019, Yuling Liu, Yunpeng Li 0006

Machine Learning (ML) based malware classifiers are vulnerable to exploitation during the training phase due to the necessity of regular retraining with samples collected from the wild. Recent studies have highlighted the efficacy of backdoor attacks in the malware domain, where attackers can manipulate the model during training by injecting samples embedded with specific triggers, causing the model to establish an association between the trigger and a designated class, thereby achieving evasion of detection. While research on backdoor attacks has been extensively explored in the field of computer vision, it has been largely overlooked in the malware domain. Unlike in the computer vision domain, the threat model in the malware domain typically restricts attackers to employing clean-label attacks (i.e., attackers do not have control over the labeling of poisoned data). However, clean-label attack methods are generally less effective compared to those that involve embedding triggers and altering sample labels to the target class (called corrupted-label attacks). To address this limitation, we propose a simple yet effective method that involves Poisoning Malware-Similar Benignware (PMSB) instead of random selection, thereby approximating the scenario of corrupted-label attacks and enhancing the effectiveness of clean-label attacks. Additionally, we introduce three similarity measurement methods based on feature-based distance, distribution-based distance, and contribution-based difference to select malware-similar benignware. Comprehensive evaluations across three different trigger types and three datasets demonstrate the superiority and general applicability of PMSB.

Haibo Liu, Chen Gong 0006, Zhenzhe Zheng 0001, Shengzhong Liu, Fan Wu 0006

Online continual learning (CL) is becoming a mainstream paradigm to learn incrementally from task streams without forgetting previously learned knowledge. However, the current online CL primarily focuses on learning performance, such as avoiding catastrophic forgetting, neglecting the critical demands of system performance, such as real-time inference. As a result, the performance of real-time inference in online CL degrades significantly due to frequent data distribution variations and time-consuming model adaptation. In this work, we propose ELITE, an online CL framework with device-cloud collaboration, to realize on-device real-time inference on time-varying task streams with performance guarantee. To realize on-device real-time inference in online CL, ELITE features a new design of the model zoo comprising various pre-trained models with the assistance of the cloud, and proposes a task-oriented on-device model selection to quickly retrieve the best-fit models instead of performing time-consuming model retraining. To prevent performance degradation on new tasks not available in the cloud, we introduces a latency-aware on-device model fine-tuning strategy to adapt to new tasks with an accuracy-latency trade-off, and dynamically updates the model zoo to enhance ELITE. Extensive evaluations on five real-world datasets have been conducted, and the results demonstrate that ELITE consistently outperforms the state-of-art solutions, improving the accuracy by 16.3% on average and reducing the response latency by up to 1.98 times.

Yunyi Zhang 0001, Ruozhen Yang, Xueqiang Xu, Rui Li 0049, Jinfeng Xiao, Jiaming Shen, Jiawei Han 0001

Hierarchical text classification aims to categorize each document into a set of classes in a label taxonomy, which is a fundamental web text mining task with broad applications such as web content analysis and semantic indexing. Most earlier works focus on fully or semi-supervised methods that require a large amount of human annotated data which is costly and time-consuming to acquire. To alleviate human efforts, in this paper, we work on hierarchical text classification with a minimal amount of supervision: using the sole class name of each node as the only supervision. Recently, large language models (LLM) have shown competitive performance on various tasks through zero-shot prompting, but this method performs poorly in the hierarchical setting because it is ineffective to include the large and structured label space in a prompt. On the other hand, previous weakly-supervised hierarchical text classification methods only utilize the raw taxonomy skeleton and ignore the rich information hidden in the text corpus that can serve as additional class-indicative features. To tackle the above challenges, we propose TELEClass, Taxonomy Enrichment and LLM-Enhanced weakly-supervised hierarchical text Classification, which combines the general knowledge of LLMs and task-specific features mined from an unlabeled corpus. TELEClass automatically enriches the raw taxonomy with class-indicative features for better label space understanding and utilizes novel LLM-based data annotation and generation methods specifically tailored for the hierarchical setting. Experiments show that TELEClass can significantly outperform previous baselines while achieving comparable performance to zero-shot prompting of LLMs with drastically less inference cost.

Chu Zhao, Enneng Yang, Yuliang Liang, Jianzhe Zhao, Guibing Guo, Xingwei Wang 0001

The distributionally robust optimization (DRO)-based graph neural network methods improve recommendation systems' out-of-distribution (OOD) generalization by optimizing the model's worst-case performance. However, these studies fail to consider the impact of noisy samples in the training data, which results in diminished generalization capabilities and lower accuracy. Through experimental and theoretical analysis, this paper reveals that current DRO-based graph recommendation methods assign greater weight to noise distribution, leading to model parameter learning being dominated by it. When the model overly focuses on fitting noise samples in the training data, it may learn irrelevant or meaningless features that cannot be generalized to OOD data. To address this challenge, we design a Distributionally Robust Graph model for OOD recommendation (DRGO). Specifically, our method first employs a simple and effective diffusion paradigm to alleviate the noisy effect in the latent space. Additionally, an entropy regularization term is introduced in the DRO objective function to avoid extreme sample weights in the worst-case distribution. Finally, we provide a theoretical proof of the generalization error bound of DRGO as well as a theoretical analysis of how our approach mitigates noisy sample effects, which helps to better understand the proposed framework from a theoretical perspective. We conduct extensive experiments on four datasets to evaluate the effectiveness of our framework against three typical distribution shifts, and the results demonstrate its superiority in both independently and identically distributed distributions (IID) and OOD.