pith. sign in

arxiv: 2510.08952 · v4 · submitted 2025-10-10 · 💻 cs.LG

When LLM Agents Meet Graph Optimization: An Automated Data Quality Improvement Approach

Zhihan Zhang , Xunkai Li , Yilong Zuo , Henan Sun , Zhenjun Li , Bing Zhou , Rong-Hua Li , Guoren Wang This is my paper

Pith reviewed 2026-05-18 08:42 UTC · model grok-4.3

classification 💻 cs.LG
keywords text-attributed graphsdata qualitymulti-agent systemslarge language modelsgraph neural networksautomated optimizationTAG quality improvement
0
0 comments X

The pith

A multi-agent LLM system automatically detects and repairs imperfections across text, structure, and labels in text-attributed graphs.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper demonstrates that both standard and LLM-enhanced graph neural networks lose accuracy when text-attributed graphs contain flaws in their text content, connections, or labels. It presents LAGA as a single framework that runs detection, planning, action, and evaluation agents in a repeating cycle to address all three flaw types at once. Earlier work handled only one flaw category with separate techniques, leaving gaps in coverage. Experiments on five datasets and sixteen baselines across nine scenarios indicate that the coordinated loop produces cleaner data and stronger downstream model results.

Core claim

LAGA formulates graph quality control as a data-centric process, integrating detection, planning, action, and evaluation agents into an automated loop that holistically enhances textual, structural, and label aspects through coordinated multi-modal optimization.

What carries the argument

LAGA, the multi-agent framework that coordinates LLM-powered agents to detect imperfections, plan repairs, execute fixes, and evaluate outcomes in a closed automated loop.

If this is right

  • Graph neural networks reach higher accuracy on data that has received coordinated fixes to text, structure, and labels.
  • A single automated loop replaces the need for separate tools that target only one type of data degradation.
  • The approach maintains gains across multiple degradation patterns and scales to additional datasets without per-scenario redesign.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If the agent loop proves stable, similar coordinated repair systems could be built for other multimodal data types such as knowledge graphs or document collections.
  • Data preparation pipelines might embed this kind of agent cycle to maintain quality continuously rather than as a one-time step.
  • Reduced reliance on manual inspection could let smaller teams run reliable graph analytics on noisy real-world sources.

Load-bearing premise

Large language model agents can reliably identify data imperfections and apply effective repairs across modalities without introducing new errors or needing extensive human oversight.

What would settle it

Apply LAGA to graphs with injected, known imperfections and measure whether downstream GNN accuracy stays the same or drops instead of rising.

Figures

Figures reproduced from arXiv: 2510.08952 by Bing Zhou, Guoren Wang, Henan Sun, Rong-Hua Li, Xunkai Li, Yilong Zuo, Zhenjun Li, Zhihan Zhang.

Figure 1
Figure 1. Figure 1: Performance comparison of two GNN backbones (GCN, GAT) and two LLM-GNN backbones (TAPE, ENGINE) across [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: The framework of LAGA, including the overall workflow of LAGA and the internal module details of each agent. [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Node clustering NMI comparison across different scenarios with perturbation ratio = 0.4. The "Org" denotes the [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Performance of LAGA with different backbones across nine scenarios, showing the accuracy improvements over the [PITH_FULL_IMAGE:figures/full_fig_p010_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: The sensitivity of LAGA to different hyperparameters [PITH_FULL_IMAGE:figures/full_fig_p011_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: The performance of LAGA under different composite scenarios. In total, four composite scenarios are considered, and [PITH_FULL_IMAGE:figures/full_fig_p011_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Expert evaluation results under nine scenarios, where ”Graph-Bef” denotes the graph before optimization, ”Graph-Aft” [PITH_FULL_IMAGE:figures/full_fig_p012_7.png] view at source ↗
read the original abstract

Text-attributed graphs (TAGs) have become a key form of graph-structured data in modern data management and analytics, combining structural relationships with rich textual semantics for diverse applications. However, the effectiveness of analytical models, particularly graph neural networks (GNNs), is highly sensitive to data quality. Our empirical analysis shows that both conventional and LLM-enhanced GNNs degrade notably under textual, structural, and label imperfections, underscoring TAG quality as a key bottleneck for reliable analytics. Existing studies have explored data-level optimization for TAGs, but most focus on specific degradation types and target a single aspect like structure or label, lacking a systematic and comprehensive perspective on data quality improvement. To address this gap, we propose LAGA (Large Language and Graph Agent), a unified multi-agent framework for comprehensive TAG quality optimization. LAGA formulates graph quality control as a data-centric process, integrating detection, planning, action, and evaluation agents into an automated loop. It holistically enhances textual, structural, and label aspects through coordinated multi-modal optimization. Extensive experiments on 5 datasets and 16 baselines across 9 scenarios demonstrate the effectiveness, robustness and scalability of LAGA, confirming the importance of data-centric quality optimization for reliable TAG analytics.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper introduces LAGA, a multi-agent framework that formulates TAG quality control as an automated data-centric process. It integrates detection, planning, action, and evaluation agents into a coordinated loop to holistically optimize textual, structural, and label quality in text-attributed graphs. The central claim is that this approach mitigates performance degradation in GNNs under imperfections and outperforms 16 baselines across 5 datasets and 9 scenarios, demonstrating effectiveness, robustness, and scalability.

Significance. If the empirical claims hold with proper controls and ablations, the work is significant for shifting emphasis toward automated, multi-modal data quality optimization in graph learning. It addresses a documented bottleneck where both conventional and LLM-enhanced GNNs degrade under TAG imperfections, and the engineering of LLM agents for coordinated graph repair could influence practical data curation pipelines in analytics and management applications.

major comments (2)
  1. [Abstract and Experiments] Abstract and §4 (Experiments): The abstract states gains over 16 baselines on 5 datasets but provides no quantitative metrics, error bars, ablation results, or details on how imperfections were introduced and measured. Without these, the claim that the multi-agent loop delivers holistic improvement cannot be evaluated for magnitude, consistency, or causality versus generic LLM prompting.
  2. [Framework] §3 (Framework): The weakest assumption—that LLM agents reliably detect and repair imperfections across modalities without introducing new errors—is load-bearing for the automated-loop claim. No failure-mode analysis, human validation of repairs, or sensitivity to coordination thresholds is referenced, leaving open whether the framework requires extensive domain tuning.
minor comments (2)
  1. [Figure 1 and Framework] Figure 1 and §3.1: The agent roles and information flow could be clarified with explicit input/output specifications to avoid ambiguity in how multi-modal signals are passed between agents.
  2. [Related Work] Related Work: A few recent LLM-based graph cleaning papers appear under-cited; adding them would better position the novelty of the coordinated agent loop.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback and the recommendation for major revision. We address each major comment point by point below and describe the revisions we will incorporate.

read point-by-point responses

  1. Referee: [Abstract and Experiments] Abstract and §4 (Experiments): The abstract states gains over 16 baselines on 5 datasets but provides no quantitative metrics, error bars, ablation results, or details on how imperfections were introduced and measured. Without these, the claim that the multi-agent loop delivers holistic improvement cannot be evaluated for magnitude, consistency, or causality versus generic LLM prompting.

    Authors: We agree that the abstract would be strengthened by including key quantitative results. In the revised manuscript we will update the abstract to report specific average performance gains (with standard deviations) across the 9 scenarios. Section 4.1 already details the imperfection injection protocols (textual noise via synonym replacement and truncation, structural edge deletion/addition at controlled rates, and label flipping) and the metrics used to quantify degradation. To directly address causality versus generic LLM prompting, we will add a new ablation study in §4 comparing the full coordinated LAGA loop against a baseline that uses the same LLM for isolated detection and repair without the planning and evaluation agents. These additions will make the magnitude, consistency, and incremental benefit of the multi-agent design explicit. revision: yes

  2. Referee: [Framework] §3 (Framework): The weakest assumption—that LLM agents reliably detect and repair imperfections across modalities without introducing new errors—is load-bearing for the automated-loop claim. No failure-mode analysis, human validation of repairs, or sensitivity to coordination thresholds is referenced, leaving open whether the framework requires extensive domain tuning.

    Authors: We acknowledge that the reliability of the agent loop is a central assumption. The evaluation agent is explicitly designed to score repairs and trigger re-planning when quality does not improve, which provides an internal safeguard against new errors. Nevertheless, we agree that additional empirical support is warranted. In the revision we will insert a dedicated failure-mode subsection that catalogs representative cases in which an action introduced a new imperfection and how the subsequent evaluation step corrected or mitigated it. We will also report human validation accuracy on a random sample of 200 repairs (stratified across the five datasets) and include a sensitivity analysis varying the coordination threshold that decides whether to accept an action or iterate. Our current results already show consistent gains across five heterogeneous datasets without per-dataset prompt engineering or fine-tuning, suggesting limited domain tuning is needed; we will make this point more explicit in the revised discussion. revision: yes

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper presents LAGA as an engineering framework that integrates detection, planning, action, and evaluation agents into an automated loop for TAG quality optimization. The abstract and provided text contain no equations, fitted parameters, or self-referential definitions that reduce the claimed holistic improvements to inputs by construction. Effectiveness is asserted via external experiments across 5 datasets and 16 baselines rather than any closed mathematical derivation or self-citation chain. The contribution is therefore self-contained as a proposed system with empirical support.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 1 invented entities

The framework rests on the untested premise that current LLMs possess sufficient semantic understanding and editing reliability for graph data; several new agent components are introduced without independent falsifiable evidence outside the reported experiments.

free parameters (1)
  • agent coordination thresholds
    Parameters controlling when detection triggers planning or when evaluation accepts a repair are likely required but not specified.
axioms (1)
  • domain assumption Large language models can accurately detect and correct textual, structural, and label imperfections in graphs
    Invoked when the detection, planning, and action agents are assumed to perform their roles effectively.
invented entities (1)
  • Detection agent, Planning agent, Action agent, Evaluation agent no independent evidence
    purpose: Form the coordinated multi-agent loop for quality control
    New components introduced by the paper; no independent evidence of their standalone performance is provided.

pith-pipeline@v0.9.0 · 5769 in / 1400 out tokens · 65626 ms · 2026-05-18T08:42:54.557059+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Integrating Graphs, Large Language Models, and Agents: Reasoning and Retrieval

    cs.AI 2026-04 unverdicted novelty 6.0

    A structured survey organizing graph-LLM integration methods by purpose, modality, and strategy across application domains.

Reference graph

Works this paper leans on

61 extracted references · 61 canonical work pages · cited by 1 Pith paper · 7 internal anchors

  1. [1]

    LAGA Technical Report

    2025. LAGA Technical Report. https://anonymous.4open.science/r/LAGA-main- FB43

    work page 2025

  2. [2]

    Wendong Bi, Lun Du, Qiang Fu, Yanlin Wang, Shi Han, and Dongmei Zhang

    work page

  3. [3]

    Make heterophilic graphs better fit gnn: A graph rewiring approach.IEEE Transactions on Knowledge and Data Engineering(2024)

    work page 2024

  4. [4]

    Vincent D Blondel, Jean-Loup Guillaume, Renaud Lambiotte, and Etienne Lefeb- vre. 2008. Fast unfolding of communities in large networks.Journal of Statistical Mechanics: Theory and Experiment2008, 10 (2008), P10008

    work page 2008

  5. [5]

    Junyu Chen, Qianqian Xu, Zhiyong Yang, Xiaochun Cao, and Qingming Huang

    work page

  6. [6]

    InProceedings of the 30th ACM International Conference on Multimedia

    A unified framework against topology and class imbalance. InProceedings of the 30th ACM International Conference on Multimedia. 180–188

    work page

  7. [7]

    Yu Chen, Lingfei Wu, and Mohammed Zaki. 2020. Iterative deep graph learning for graph neural networks: Better and robust node embeddings.Advances in neural information processing systems33 (2020), 19314–19326

    work page 2020

  8. [8]

    Zhikai Chen, Haitao Mao, Jingzhe Liu, Yu Song, Bingheng Li, Wei Jin, Bahare Fatemi, Anton Tsitsulin, Bryan Perozzi, Hui Liu, et al. 2024. Text-space Graph Foundation Models: Comprehensive Benchmarks and New Insights.arXiv preprint arXiv:2406.10727(2024)

    work page arXiv 2024

  9. [9]

    Enyan Dai, Charu Aggarwal, and Suhang Wang. 2021. Nrgnn: Learning a label noise resistant graph neural network on sparsely and noisily labeled graphs. In Proceedings of the 27th ACM SIGKDD conference on knowledge discovery & data mining. 227–236

    work page 2021

  10. [10]

    Sijie Dong, Soror Sahri, Themis Palpanas, and Qitong Wang. 2025. Automated Data Quality Validation in an End-to-End GNN Framework.arXiv preprint arXiv:2502.10667(2025)

    work page arXiv 2025

  11. [11]

    Enjun Du, Xunkai Li, Tian Jin, Zhihan Zhang, Rong-Hua Li, and Guoren Wang

    work page

  12. [12]

    Graphmaster: Automated graph synthesis via llm agents in data-limited environments.arXiv preprint arXiv:2504.00711(2025)

    work page arXiv 2025

  13. [13]

    Xuefeng Du, Tian Bian, Yu Rong, Bo Han, Tongliang Liu, Tingyang Xu, Wenbing Huang, Yixuan Li, and Junzhou Huang. 2021. Noise-robust graph learning by estimating and leveraging pairwise interactions.arXiv preprint arXiv:2106.07451 (2021)

    work page arXiv 2021

  14. [14]

    Will Hamilton, Zhitao Ying, and Jure Leskovec. 2017. Inductive representation learning on large graphs.Advances in neural information processing systems30 (2017)

    work page 2017

  15. [15]

    Xiaoxin He, Xavier Bresson, Thomas Laurent, Adam Perold, Yann LeCun, and Bryan Hooi. 2023. Harnessing explanations: Llm-to-lm interpreter for enhanced text-attributed graph representation learning.arXiv preprint arXiv:2305.19523 (2023)

    work page arXiv 2023

  16. [16]

    Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. 2021. LoRA: Low-Rank Adaptation of Large Language Models.arXiv preprint arXiv:2106.09685(2021)

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  17. [17]

    Yuchuan Huang and Mohamed F Mokbel. 2023. Sparcle: Boosting the Accu- racy of Data Cleaning Systems through Spatial Awareness.arXiv preprint arXiv:2311.04836(2023)

    work page arXiv 2023

  18. [18]

    Jinhao Jiang, Kun Zhou, Wayne Xin Zhao, Yang Song, Chen Zhu, Hengshu Zhu, and Ji-Rong Wen. 2024. Kg-agent: An efficient autonomous agent framework for complex reasoning over knowledge graph.arXiv preprint arXiv:2402.11163 (2024)

    work page arXiv 2024

  19. [19]

    Wei Jin, Yao Ma, Xiaorui Liu, Xianfeng Tang, Suhang Wang, and Jiliang Tang

    work page

  20. [20]

    InProceedings of the 26th ACM SIGKDD international conference on knowledge discovery & data mining

    Graph structure learning for robust graph neural networks. InProceedings of the 26th ACM SIGKDD international conference on knowledge discovery & data mining. 66–74

    work page

  21. [21]

    Mingxuan Ju, Tong Zhao, Wenhao Yu, Neil Shah, and Yanfang Ye. 2023. Graph- patcher: mitigating degree bias for graph neural networks via test-time augmen- tation.Advances in Neural Information Processing Systems36 (2023), 55785–55801

    work page 2023

  22. [22]

    Jian Kang, Yan Zhu, Yinglong Xia, Jiebo Luo, and Hanghang Tong. 2022. Rawls- gcn: Towards rawlsian difference principle on graph convolutional network. In Proceedings of the ACM Web Conference 2022. 1214–1225

    work page 2022

  23. [23]

    Thomas N Kipf and Max Welling. 2016. Semi-supervised classification with graph convolutional networks.arXiv preprint arXiv:1609.02907(2016)

    work page internal anchor Pith review Pith/arXiv arXiv 2016

  24. [24]

    Junseok Lee, Yunhak Oh, Yeonjun In, Namkyeong Lee, Dongmin Hyun, and Chanyoung Park. 2022. Grafn: Semi-supervised node classification on graph with few labels via non-parametric distribution assignment. InProceedings of the 45th International ACM SIGIR Conference on Research and Development in Information Retrieval. 2243–2248

    work page 2022

  25. [25]

    Wen-Zhi Li, Chang-Dong Wang, Hui Xiong, and Jian-Huang Lai. 2023. Graphsha: Synthesizing harder samples for class-imbalanced node classification. InProceed- ings of the 29th ACM SIGKDD conference on knowledge discovery and data mining. 1328–1340

    work page 2023

  26. [26]

    Xianxian Li, Qiyu Li, Haodong Qian, Jinyan Wang, et al . 2024. Contrastive learning of graphs under label noise.Neural networks172 (2024), 106113

    work page 2024

  27. [27]

    Xunkai Li, Zhengyu Wu, Jiayi Wu, Hanwen Cui, Jishuo Jia, Rong-Hua Li, and Guoren Wang. 2024. Graph Learning in the Era of LLMs: A Survey from the Perspective of Data, Models, and Tasks.arXiv preprint arXiv:2412.12456(2024)

    work page arXiv 2024

  28. [28]

    Nian Liu, Xiao Wang, Lingfei Wu, Yu Chen, Xiaojie Guo, and Chuan Shi. 2022. Compact graph structure learning via mutual information compression. InPro- ceedings of the ACM web conference 2022. 1601–1610

    work page 2022

  29. [29]

    Yixin Liu, Yu Zheng, Daokun Zhang, Hongxu Chen, Hao Peng, and Shirui Pan

    work page

  30. [30]

    InProceedings of the ACM Web Conference 2022

    Towards unsupervised deep graph structure learning. InProceedings of the ACM Web Conference 2022. 1392–1403

    work page 2022

  31. [31]

    Zemin Liu, Trung-Kien Nguyen, and Yuan Fang. 2021. Tail-gnn: Tail-node graph neural networks. InProceedings of the 27th ACM SIGKDD conference on knowledge discovery & data mining. 1109–1119

    work page 2021

  32. [32]

    Péter Mernyei and Cătălina Cangea. 2020. Wiki-cs: A wikipedia-based benchmark for graph neural networks.arXiv preprint arXiv:2007.02901(2020)

    work page arXiv 2020

  33. [33]

    Makbule Gulcin Ozsoy, Leila Messallem, Jon Besga, and Gianandrea Minneci

    work page

  34. [34]

    Text2cypher: Bridging natural language and graph databases.arXiv preprint arXiv:2412.10064(2024)

    work page arXiv 2024

  35. [35]

    Jiawen Qin, Haonan Yuan, Qingyun Sun, Lyujin Xu, Jiaqi Yuan, Pengfeng Huang, Zhaonan Wang, Xingcheng Fu, Hao Peng, Jianxin Li, et al . 2024. Igl-bench: Establishing the comprehensive benchmark for imbalanced graph learning.arXiv preprint arXiv:2406.09870(2024)

    work page arXiv 2024

  36. [36]

    Nils Reimers and Iryna Gurevych. 2019. Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks. InProceedings of the 2019 Conference on Empirical Methods in Natural Language Processing (EMNLP). Association for Computational Linguistics, Hong Kong, China

    work page 2019

  37. [37]

    Gerard Salton and Christopher Buckley. 1988. Term-weighting approaches in automatic text retrieval.Information Processing & Management24, 5 (1988), 513–523

    work page 1988

  38. [38]

    Rubab Zahra Sarfraz. 2024. Towards Semi-Supervised Data Quality Detection in Graphs.Proceedings of the VLDB Endowment. ISSN2150 (2024), 8097

    work page 2024

  39. [39]

    Sebastian Schelter, Dustin Lange, Philipp Schmidt, Meltem Celikel, Felix Biess- mann, and Andreas Grafberger. 2018. Automating large-scale data quality verifi- cation.Proceedings of the VLDB Endowment11, 12 (2018), 1781–1794

    work page 2018

  40. [40]

    Phanwadee Sinthong, Dhaval Patel, Nianjun Zhou, Shrey Shrivastava, Arun Iyengar, and Anuradha Bhamidipaty. 2021. DQDF: data-quality-aware dataframes. Proceedings of the VLDB Endowment15, 4 (2021), 949–957

    work page 2021

  41. [41]

    Yifu Sun and Haoming Jiang. 2019. Contextual text denoising with masked language models.arXiv preprint arXiv:1910.14080(2019)

    work page arXiv 2019

  42. [42]

    Gemma Team, Aishwarya Kamath, Johan Ferret, Shreya Pathak, Nino Vieillard, Ramona Merhej, Sarah Perrin, Tatiana Matejovicova, Alexandre Ramé, Morgane Rivière, et al. 2025. Gemma 3 technical report.arXiv preprint arXiv:2503.19786 (2025)

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  43. [43]

    Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timothée Lacroix, Baptiste Rozière, Naman Goyal, Eric Hambro, Faisal Azhar, et al. 2023. Llama: Open and efficient foundation language models.arXiv preprint arXiv:2302.13971(2023)

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  44. [44]

    Jana Vatter, Maurice L Rochau, Ruben Mayer, and Hans-Arno Jacobsen. [n.d.]. Experiment & Benchmark Paper: To What Extent Does Quality Matter? The Impact of Graph Data Quality on GNN Model Performance.Proceedings of the VLDB Endowment. ISSN2150 ([n. d.]), 8097

    work page

  45. [45]

    Petar Veličković, Guillem Cucurull, Arantxa Casanova, Adriana Romero, Pietro Lio, and Yoshua Bengio. 2017. Graph attention networks.arXiv preprint arXiv:1710.10903(2017)

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  46. [46]

    Liang Wang, Wei Zhao, Ruoyu Jia, Sujian Li, and Jingming Liu. 2019. Denoising based sequence-to-sequence pre-training for text generation.arXiv preprint arXiv:1908.08206(2019)

    work page arXiv 2019

  47. [47]

    Zhonghao Wang, Danyu Sun, Sheng Zhou, Haobo Wang, Jiapei Fan, Longtao Huang, and Jiajun Bu. 2024. Noisygl: A comprehensive benchmark for graph neural networks under label noise.Advances in Neural Information Processing Systems37 (2024), 38142–38170

    work page 2024

  48. [48]

    Tian Xie, Bin Wang, and C-C Jay Kuo. 2022. Graphhop: An enhanced label propagation method for node classification.IEEE Transactions on Neural Networks and Learning Systems34, 11 (2022), 9287–9301

    work page 2022

  49. [49]

    Yao Xu, Shizhu He, Jiabei Chen, Zihao Wang, Yangqiu Song, Hanghang Tong, Guang Liu, Kang Liu, and Jun Zhao. 2024. Generate-on-graph: Treat llm as both agent and kg in incomplete knowledge graph question answering.arXiv preprint arXiv:2404.14741(2024)

    work page arXiv 2024

  50. [50]

    Hao Yan, Chaozhuo Li, Ruosong Long, Chao Yan, Jianan Zhao, Wenwen Zhuang, Jun Yin, Peiyan Zhang, Weihao Han, Hao Sun, et al . 2023. A comprehensive study on text-attributed graphs: Benchmarking and rethinking.Advances in Neural Information Processing Systems36 (2023), 17238–17264

    work page 2023

  51. [51]

    An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, et al. 2025. Qwen3 technical report.arXiv preprint arXiv:2505.09388(2025)

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  52. [52]

    Sukwon Yun, Kibum Kim, Kanghoon Yoon, and Chanyoung Park. 2022. Lte4g: Long-tail experts for graph neural networks. InProceedings of the 31st ACM International Conference on Information & Knowledge Management. 2434–2443

    work page 2022

  53. [53]

    Zhihan Zhang, Xunkai Li, Zhu Lei, Guang Zeng, Ronghua Li, and Guoren Wang

    work page

  54. [54]

    Rethinking Graph Structure Learning in the Era of LLMs.arXiv preprint arXiv:2503.21223(2025). 13

    work page arXiv 2025

  55. [55]

    Zihao Zhang, Xunkai Li, Rong-Hua Li, Bing Zhou, Zhenjun Li, and Guoren Wang. 2025. Toward General and Robust LLM-enhanced Text-attributed Graph Learning.arXiv preprint arXiv:2504.02343(2025)

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  56. [56]

    Zhongjian Zhang, Xiao Wang, Huichi Zhou, Yue Yu, Mengmei Zhang, Cheng Yang, and Chuan Shi. 2024. Can Large Language Models Improve the Adversarial Robustness of Graph Neural Networks?arXiv preprint arXiv:2408.08685(2024)

    work page arXiv 2024

  57. [57]

    Yonghua Zhu, Lei Feng, Zhenyun Deng, Yang Chen, Robert Amor, and Michael Witbrock. 2024. Robust node classification on graph data with graph and label noise. InProceedings of the AAAI conference on artificial intelligence, Vol. 38. 17220–17227

    work page 2024

  58. [58]

    Yun Zhu, Yaoke Wang, Haizhou Shi, and Siliang Tang. 2024. Efficient tuning and inference for large language models on textual graphs.arXiv preprint arXiv:2401.15569(2024)

    work page arXiv 2024

  59. [59]

    Yanqiao Zhu, Weizhi Xu, Jinghao Zhang, Qiang Liu, Shu Wu, and Liang Wang

    work page

  60. [60]

    Deep graph structure learning for robust representations: A survey.arXiv preprint arXiv:2103.0303614 (2021), 1–1

    work page arXiv 2021

  61. [61]

    Dongcheng Zou, Hao Peng, Xiang Huang, Renyu Yang, Jianxin Li, Jia Wu, Chun- yang Liu, and Philip S Yu. 2023. Se-gsl: A general and effective graph structure learning framework through structural entropy optimization. InProceedings of the ACM Web Conference 2023. 499–510. 14

    work page 2023