Pith. sign in

REVIEW 4 major objections 6 minor 44 references

Graph Foundation Models for Recommendation: A Comprehensive Survey

T0 review · 4 major / 6 minor · reviewed 2026-08-08 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read This survey claims to be the first comprehensive overview of graph foundation model recommender systems and organizes the field into a three-branch taxonomy.

desk verdict A useful taxonomy of GNN+LLM hybrid recommenders, but the claim to be the first overview of graph foundation models doesn't hold up against the paper's own definition. read the letter →

arxiv 2502.08346 v3 pith:OCRM7XHB submitted 2025-02-12 cs.IR cs.AIcs.LG

Bin Wu , Yihang Wang , Yuanhao Zeng , Jiawei Liu , Jiashu Zhao , Cheng Yang , Yawen Li , Long Xia
show 2 more authors
Dawei Yin Chuan Shi
This is my paper · ORCID
classification cs.IRcs.AIcs.LG
keywords graphfoundationmodelsrecommendersystemsneuralnetworkslargelanguagetaxonomyofGFM-basedLLM-graphharmonization
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

Recommender systems process two kinds of data: user-item interaction graphs and textual descriptions. Graph neural networks capture the graph side well, and large language models capture the text side, but each fails at the other. This survey argues that graph foundation models, which combine GNNs and LLMs, are becoming a distinct paradigm for recommendation, and it claims to be the first overview to organize this emerging field. Its proposed taxonomy groups all current methods into three strategies: graph-augmented LLM, LLM-augmented graph, and LLM-graph harmonization. A reader would care because the taxonomy gives researchers a shared map for comparing methods and locating open problems.

What carries the argument

The organizing device is the taxonomy in Section 2.3 and Figure 2. It classifies every GFM-based recommender by the direction of information flow between the graph and the language model: Graph-Augmented LLM (with token-level infusion and context-level infusion), LLM-Augmented Graph (with topology augmentation and feature augmentation), and LLM-Graph Harmonization (with embedding fusion and embedding alignment). This three-branch structure does the argument's work: once a method is placed in a branch, its core mechanism, interface design, and known trade-offs are supposed to follow from the branch's definition.

What would settle it

A concrete test is to take every method named in the paper's Figure 2 and check each against all three branch definitions; if even one method satisfies two definitions simultaneously, the taxonomy is not exclusive. A published GFM-based recommender that the paper omits and that cannot be placed in any branch would similarly falsify the claim that the map is complete.

Watch

Extended reading notes

Core claim

The paper's central claim is that graph foundation models for recommender systems form a coherent field whose methods can be fully described by a three-way taxonomy based on which component leads the integration. In graph-augmented LLM methods, structural information from graphs is fed into an LLM, either as special tokens or as generated context, so that the language model's reasoning drives recommendation. In LLM-augmented graph methods, the graph is the lead: the LLM adds edges, nodes, or improved features, and a GNN does the recommendation. In LLM-graph harmonization, text embeddings from an LLM and structural embeddings from a GNN are combined on equal footing through fusion or alignment. The survey further splits each branch into subcategories and presents them as an exhaustive map of the current literature.

Load-bearing premise

The load-bearing premise is that every method in the field fits cleanly into exactly one of the three taxonomy branches, with no method needing two categories or a new one.

Editorial extensions

If this is right

  • New GFM-based recommendation methods can be positioned in one of the three branches, giving the field a common vocabulary for describing contributions.
  • Researchers can identify under-explored combinations, such as the relatively small harmonization branch, and target them deliberately.
  • The taxonomy implies that the choice of integration point, tokens, context, topology, features, or embeddings, is the main design decision in this area.
  • The survey's listed challenges, computational cost, robustness, multi-modal fusion, end-to-end optimization, and knowledge-preference gap, become a concrete agenda for the next generation of models.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • We infer that the taxonomy's exclusivity is already strained by boundary cases that the paper itself discusses, such as LLMGR, which both injects graph embeddings as tokens and changes the LLM's output layer; such methods could plausibly live in two branches.
  • We infer that because most cited methods are recent preprints, the survey is a snapshot of a fast-moving field, and the taxonomy will likely need new branches or subcategories as hybrid methods accumulate.
  • We infer that the 'knowledge-preference gap' challenge points toward a testable research program: measuring whether LLM-generated neutral descriptions underperform more engaging text in online recommendation experiments.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 6 minor

Summary. This survey reviews recent recommender systems that combine graph neural networks (GNNs) with large language models (LLMs), referring to these combined models as graph foundation models (GFMs). The paper organizes the field into a three-branch taxonomy: Graph-Augmented LLM, LLM-Augmented Graph, and LLM-Graph Harmonization, with further subcategories such as token-level versus context-level infusion, topology versus feature augmentation, and embedding fusion versus embedding alignment. It describes roughly thirty methods within these categories, illustrates the taxonomy in Figures 2–5, and closes with challenges and future directions. The central claim, stated in the abstract and conclusion, is that this is the first comprehensive overview of GFM-based recommender systems with a logically organized taxonomy.

Significance. If the scope and taxonomy are taken at face value, the paper is a useful and timely map of an emerging intersection between graph learning and LLMs for recommendation. The three-way distinction is intuitive, the figures are clear, and the coverage of recent 2023–2025 work is broad. The survey also gives a fair account of trade-offs within each branch, such as plug-and-play flexibility of LLM-augmented graph methods versus their limited use of world knowledge. The main value is organizational rather than technical, since the paper contains no experiments or new methods. The significance is therefore contingent on the taxonomy being both faithful to the surveyed methods and aligned with the paper's own definition of GFMs; that alignment is presently not established.

major comments (4)
  1. [§2.2, §3–§5] The definition in §2.2 restricts GFMs to models that are 'pre-trained on large datasets' and 'incorporate graph structures,' acquiring 'emergence and homogenization during pre-training.' However, many methods surveyed in Sections 3–5 are not described as graph-pretrained. For example, LLMRG (§4.1) builds inference and divergence graphs with an LLM and then trains a GNN for the recommendation task; CSRec, FineRec, SAGCN, and LLM-KERec (§4.1) extract relations or opinions with LLMs and train GNNs on the target data; GaCLLM (§4.2) replaces GNN message passing with LLM-based textual message passing; and DynLLM, LKPNR, DALR, LLMRec, and RLMRec (§5) fuse or align GNN and LLM embeddings without any stated graph pre-training phase. As written, the paper either uses a much broader notion of GFM than the one it defines, or it surveys methods that are not GFMs under its own definition. This mismatch directly affects the abstract and conclusion's claim of presenting 'the first comprehensive overview of GFM-based RS.' The authors should either broaden the definition of GFM in §2.2 to explicitly include GNN+LLM hybrids trained without graph pre-training, or re-scope the title and claims to 'GNN+LLM hybrid recommender systems' and adjust the set of included methods accordingly.
  2. [§5, opening paragraph] The paragraph states that 'this section introduces a novel framework that optimally balances computational efficiency, adaptability, and reasoning capabilities,' but the section then reviews existing harmonization methods such as DynLLM, LKPNR, DALR, LLMRec, and RLMRec. No new framework is introduced. If 'framework' is meant to refer to the embedding fusion versus embedding alignment categorization, the sentence should say so explicitly and avoid implying that the harmonization approach itself is a novel contribution of this paper.
  3. [§4.1, Edge-level Expansion] The sentence 'In addition to mapping edges based on semantic similarity, connections can also be established based on the similarity' is incomplete and does not specify what kind of similarity is meant. This makes the methodological description of edge construction imprecise at exactly the point where the survey is supposed to clarify how LLM-generated relationships are turned into graph edges. The sentence should be completed with the intended quantity, for example embedding similarity or vector similarity, and ideally with a citation to the works that use it.
  4. [Figure 2 and §2.3] The taxonomy is presented as a clean three-way partition, but several surveyed methods already span the proposed categories. LLMRG is listed under both Edge-Level Expansion and Node-Level Expansion in Figure 2, and LLMGR (§3.1) is described not only as injecting graph embeddings into the token sequence but also as modifying the LLM's output layer. The paper does not discuss how to assign methods that combine multiple strategies, despite claiming that the taxonomy allows future work to be 'easily categorized.' A short paragraph stating assignment rules or a primary-category criterion would make the taxonomy more robust and prevent the 'comprehensive' claim from resting on an imprecise classification principle.
minor comments (6)
  1. [§4.2] The sentence 'directly improving the node features in the graph without altering the topological structure is an task where LLMs truly excel' contains a grammatical typo: 'an task' should be 'a task.'
  2. [§1 and Figure 2] The naming of the second taxonomy branch is inconsistent: the text uses 'LLM-augmented graph' while Figure 2 uses 'LLM–Augmented Graph.' Please standardize the capitalization and hyphenation.
  3. [§5.3] The sentence 'embedding alignment which is sensitive to noise depends on high-quality training data' needs a comma after 'noise' for readability.
  4. [§6] The word 'Contrarily' is informal and slightly ambiguous; 'In contrast' or 'Conversely' would be clearer in the paragraph on end-to-end optimization.
  5. [References] The reference '[Runfeng et al., 2023]' uses a given-name ordering that is inconsistent with the other references and may be indexed incorrectly; the entry should be checked against the original paper's author list.
  6. [Introduction and Conclusion] The paper does not state any inclusion criteria, search process, or time window for the surveyed literature. Since the central claim is 'comprehensive overview,' a brief methodology paragraph describing how papers were selected and which venues or preprint sources were covered would strengthen the claim.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: survey taxonomy is an organizational contribution independent of its cited definition.

full rationale

This is a survey paper, not a derivation. It organizes existing methods into a taxonomy and does not contain equations, fitted parameters, predictions, or first-principles results. The only self-citation is the definition of graph foundation models taken from the authors' prior survey [Liu et al., 2023b], cited in Section 2.2 as the source of the term and its characterization. That citation is used as a standard definition rather than as load-bearing evidence for any conclusion; the taxonomy in Section 2.3 is constructed from the surveyed methods themselves and does not depend on the prior survey's validity. The abstract's 'first comprehensive overview' claim is a novelty assertion, not a circular derivation. A potential scope mismatch (many surveyed methods do not obviously satisfy the Section 2.2 pre-training requirement) is a correctness or coverage concern, not a circularity concern. No circular step can be exhibited, so the appropriate finding is no significant circularity.

Assumptions & free parameters 0 free parameters · 2 assumptions · 0 invented entities

Survey paper; no free parameters or invented entities. The taxonomy is the paper's contribution, not a free parameter.

assumptions (2)
  • domain assumption The reviewed methods can be unambiguously assigned to exactly one of the three taxonomy categories.
    Underlies the paper's central contribution; methods like LLMGR are placed under Graph-Augmented LLM, but some methods combine multiple strategies, which the paper does not discuss.
  • domain assumption The descriptions of cited methods accurately reflect the original papers.
    The survey's value depends on faithful representation of works that the authors did not evaluate experimentally.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Graph Foundation Models for Recommendation: A Comprehensive Survey." pith.science (2026). https://pith.science/paper/OCRM7XHB

@misc{pith2026250208346,
  author       = {Pith},
  title        = {Pith review of: Graph Foundation Models for Recommendation: A Comprehensive Survey},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OCRM7XHB}},
  note         = {Machine review of arXiv:2502.08346}
}
read the original abstract

Recommender systems (RS) serve as a fundamental tool for navigating the vast expanse of online information, with deep learning advancements playing an increasingly important role in improving ranking accuracy. Among these, graph neural networks (GNNs) excel at extracting higher-order structural information, while large language models (LLMs) are designed to process and comprehend natural language, making both approaches highly effective and widely adopted. Recent research has focused on graph foundation models (GFMs), which integrate the strengths of GNNs and LLMs to model complex RS problems more efficiently by leveraging the graph-based structure of user-item relationships alongside textual understanding. In this survey, we provide a comprehensive overview of GFM-based RS technologies by introducing a clear taxonomy of current approaches, diving into methodological details, and highlighting key challenges and future directions. By synthesizing recent advancements, we aim to offer valuable insights into the evolving landscape of GFM-based recommender systems.

Figures

Figures reproduced from arXiv: 2502.08346 by the authors.

Figure 1
Figure 1. An overview of GFM-based RS. Compared with GNN [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. A taxonomy of GFM-based recommender systems. [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. The illustration of graph-augmented LLM methods: a) [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: The illustration of LLM-augmented graph methods: a) [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]
Figure 5
Figure 5. Figure 5: The illustration of LLM-graph harmonization methods: a) [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

44 extracted references · 26 canonical work pages

  1. [1]

    Knowledge Graphs as Context Sources for LLM-Based Explanations of Learning Recommendations

    [Abu-Rasheed et al., 2024] Hasan Abu-Rasheed, Christian Weber, and Madjid Fathi. Knowledge graphs as context sources for llm-based explanations of learning recommen- dations. arXiv preprint arXiv:2403.03008,

    work page Pith review arXiv 2024

  2. [4]

    Leverage knowledge graph and large language model for law article recommendation: A case study of chinese criminal law

    [Chen et al., 2024] Yongming Chen, Miner Chen, Ye Zhu, et al. Leverage knowledge graph and large language model for law article recommendation: A case study of chinese criminal law. arXiv preprint arXiv:2410.04949,

    work page arXiv 2024

  3. [5]

    Deep neural networks for youtube recom- mendations

    [Covington et al., 2016] Paul Covington, Jay Adams, and Emre Sargin. Deep neural networks for youtube recom- mendations. In Proc. of RecSys, pages 191–198,

    work page 2016

  4. [7]

    Towards graph foundation models for personalization

    [Damianou et al., 2024] Andreas Damianou, Francesco Fab- bri, Paul Gigioli, et al. Towards graph foundation models for personalization. In Companion Proceedings of the ACM on Web Conference 2024, pages 1798–1802,

    work page 2024

  5. [8]

    Large language model with graph convolution for recom- mendation

    [Du et al., 2024] Yingpeng Du, Ziyan Wang, Zhu Sun, et al. Large language model with graph convolution for recom- mendation. arXiv preprint arXiv:2402.08859,

    arXiv 2024

  6. [9]

    A survey of graph neural networks for recommender systems: Challenges, methods, and directions

    [Gao et al., 2023] Chen Gao, Yu Zheng, Nian Li, et al. A survey of graph neural networks for recommender systems: Challenges, methods, and directions. ACM Transactions on Recommender Systems, 1(1):1–51,

    work page 2023

  7. [11]

    Integrating large language models with graphical session-based recommendation

    [Guo et al., 2024] Naicheng Guo, Hongwei Cheng, Qianqiao Liang, et al. Integrating large language models with graphical session-based recommendation. arXiv preprint arXiv:2402.16539,

    arXiv 2024

  8. [12]

    Bridg- ing the user-side knowledge gap in knowledge-aware rec- ommendations with large language models

    [Hu et al., 2024] Zheng Hu, Zhe Li, Ziyun Jiao, et al. Bridg- ing the user-side knowledge gap in knowledge-aware rec- ommendations with large language models. arXiv preprint arXiv:2412.13544,

    arXiv 2024

Show all 44 references
  1. [13]

    Topic-aware knowledge graph with large lan- guage models for interoperability in recommender systems

    [Jeon et al., 2024] Minhye Jeon, Seokho Ahn, and Young- Duk Seo. Topic-aware knowledge graph with large lan- guage models for interoperability in recommender systems. arXiv preprint arXiv:2412.20163,

    2024 arXiv

  2. [14]

    Hetgcot-rec: Heterogeneous graph-enhanced chain-of- thought llm reasoning for journal recommendation

    [Jia et al., 2025] Runsong Jia, Mengjia Wu, Ying Ding, et al. Hetgcot-rec: Heterogeneous graph-enhanced chain-of- thought llm reasoning for journal recommendation. arXiv preprint arXiv:2501.01203,

    2025

  3. [16]

    Bert: Pre-training of deep bidirectional transformers for lan- guage understanding

    [Kenton and Toutanova, 2019] Jacob Devlin Ming- Wei Chang Kenton and Lee Kristina Toutanova. Bert: Pre-training of deep bidirectional transformers for lan- guage understanding. In Proc. of NAACL-HLT, volume

    2019

  4. [18]

    Learning structure and knowledge aware representation with large language models for concept recommendation

    [Li et al., 2024] Qingyao Li, Wei Xia, Kounianhua Du, et al. Learning structure and knowledge aware representation with large language models for concept recommendation. arXiv preprint arXiv:2405.12442,

    2024 arXiv

  5. [19]

    Towards graph foundation models: A survey and beyond

    [Liu et al., 2023b] Jiawei Liu, Cheng Yang, Zhiyuan Lu, et al. Towards graph foundation models: A survey and beyond. arXiv preprint arXiv:2310.11829,

    arXiv

  6. [20]

    Triple modality fusion: Aligning visual, textual, and graph data with large language models for multi-behavior recom- mendations

    [Ma et al., 2024a] Luyi Ma, Xiaohan Li, Zezhong Fan, et al. Triple modality fusion: Aligning visual, textual, and graph data with large language models for multi-behavior recom- mendations. arXiv preprint arXiv:2410.12228,

    arXiv

  7. [21]

    Xrec: Large language models for explainable recommenda- tion

    [Ma et al., 2024b] Qiyao Ma, Xubin Ren, and Chao Huang. Xrec: Large language models for explainable recommenda- tion. arXiv preprint arXiv:2406.02377,

    arXiv

  8. [22]

    Lightlm: a lightweight deep and narrow language model for generative recommendation

    [Mei and Zhang, 2023] Kai Mei and Yongfeng Zhang. Lightlm: a lightweight deep and narrow language model for generative recommendation. arXiv preprint arXiv:2310.17488,

    2023 arXiv

  9. [23]

    Denoising alignment with large language model for rec- ommendation

    [Peng et al., 2024] Yingtao Peng, Chen Gao, Yu Zhang, et al. Denoising alignment with large language model for rec- ommendation. ACM Transactions on Information Systems,

    2024

  10. [24]

    Unveiling user preferences: A knowledge graph and llm- driven approach for conversational recommendation

    [Qiu et al., 2024] Zhangchi Qiu, Linhao Luo, Shirui Pan, et al. Unveiling user preferences: A knowledge graph and llm- driven approach for conversational recommendation. arXiv preprint arXiv:2411.14459,

    2024

  11. [25]

    Language models are unsupervised multitask learners

    [Radford et al., 2019] Alec Radford, Jeffrey Wu, Rewon Child, et al. Language models are unsupervised multitask learners. OpenAI blog, 1(8):9,

    2019

  12. [26]

    Representation learning with large language models for recommendation

    [Ren et al., 2024] Xubin Ren, Wei Wei, Lianghao Xia, et al. Representation learning with large language models for recommendation. In Proceedings of the ACM on Web Con- ference 2024, pages 3464–3475,

    2024

  13. [27]

    Lkpnr: Llm and kg for personal- ized news recommendation framework

    [Runfeng et al., 2023] Xie Runfeng, Cui Xiangyang, Yan Zhou, et al. Lkpnr: Llm and kg for personal- ized news recommendation framework. arXiv preprint arXiv:2308.12028,

    2023 arXiv

  14. [28]

    Llm is knowledge graph reasoner: Llm’s intuition-aware knowledge graph reasoning for cold-start sequential recommendation

    [Sakurai et al., 2024] Keigo Sakurai, Ren Togo, Takahiro Ogawa, et al. Llm is knowledge graph reasoner: Llm’s intuition-aware knowledge graph reasoning for cold-start sequential recommendation. arXiv preprint arXiv:2412.12464,

    2024 arXiv

  15. [29]

    An automatic graph construction framework based on large language models for recommendation

    [Shan et al., 2024] Rong Shan, Jianghao Lin, Chenxu Zhu, et al. An automatic graph construction framework based on large language models for recommendation. arXiv preprint arXiv:2412.18241,

    2024 arXiv

  16. [30]

    Exploring user retrieval integration towards large lan- guage models for cross-domain sequential recommendation

    [Shen et al., 2024] Tingjia Shen, Hao Wang, Jiaqing Zhang, et al. Exploring user retrieval integration towards large lan- guage models for cross-domain sequential recommendation. arXiv preprint arXiv:2406.03085,

    2024 arXiv

  17. [31]

    Mgat: Multimodal graph attention network for rec- ommendation

    [Tao et al., 2020] Zhulin Tao, Yinwei Wei, Xiang Wang, et al. Mgat: Multimodal graph attention network for rec- ommendation. Information Processing & Management , 57(5):102277,

    2020

  18. [33]

    Enhancing recommender systems with large language model reasoning graphs

    [Wang et al., 2023] Yan Wang, Zhixuan Chu, Xin Ouyang, et al. Enhancing recommender systems with large language model reasoning graphs. arXiv preprint arXiv:2308.10835,

    2023 arXiv

  19. [34]

    Enabling explainable recommendation in e-commerce with llm-powered product knowledge graph

    [Wang et al., 2024b] Menghan Wang, Yuchen Guo, Duan- feng Zhang, et al. Enabling explainable recommendation in e-commerce with llm-powered product knowledge graph. arXiv preprint arXiv:2412.01837,

    arXiv

  20. [35]

    Llm-enhanced user-item interactions: Lever- aging edge information for optimized recommendations

    [Wang et al., 2024d] Xinyuan Wang, Liang Wu, Liangjie Hong, et al. Llm-enhanced user-item interactions: Lever- aging edge information for optimized recommendations. arXiv preprint arXiv:2402.09617,

    arXiv

  21. [36]

    Llmrec: Large language models with graph augmentation for recommendation

    [Wei et al., 2024] Wei Wei, Xubin Ren, Jiabin Tang, et al. Llmrec: Large language models with graph augmentation for recommendation. In In Proc. of WSDM, pages 806–815,

    2024

  22. [37]

    A comprehensive survey on graph neural networks

    [Wu et al., 2020] Zonghan Wu, Shirui Pan, Fengwen Chen, et al. A comprehensive survey on graph neural networks. IEEE transactions on neural networks and learning sys- tems, 32(1):4–24,

    2020

  23. [38]

    Graph neural networks in recommender systems: a survey

    [Wu et al., 2022] Shiwen Wu, Fei Sun, Wentao Zhang, et al. Graph neural networks in recommender systems: a survey. ACM Computing Surveys, 55(5):1–37,

    2022

  24. [39]

    Palr: Personalization aware llms for recommendation

    [Yang et al., 2023] Fan Yang, Zheng Chen, Ziyan Jiang, et al. Palr: Personalization aware llms for recommendation. arXiv preprint arXiv:2305.07622,

    2023 arXiv

  25. [40]

    Ac- tions speak louder than words: Trillion-parameter sequen- tial transducers for generative recommendations

    [Zhai et al., 2024] Jiaqi Zhai, Lucy Liao, Xing Liu, et al. Ac- tions speak louder than words: Trillion-parameter sequen- tial transducers for generative recommendations. In Proc. of ICML,

    2024

  26. [41]

    Ro- bust recommender system: a survey and future directions

    [Zhang et al., 2023] Kaike Zhang, Qi Cao, Fei Sun, et al. Ro- bust recommender system: a survey and future directions. arXiv preprint arXiv:2309.02057,

    2023 arXiv

  27. [42]

    Finerec: Exploring fine-grained sequential recommenda- tion

    [Zhang et al., 2024] Xiaokun Zhang, Bo Xu, Youlin Wu, et al. Finerec: Exploring fine-grained sequential recommenda- tion. In Proc. of SIGIR, pages 1599–1608,

    2024

  28. [43]

    A survey of large language models

    [Zhao et al., 2023] Wayne Xin Zhao, Kun Zhou, Junyi Li, et al. A survey of large language models. arXiv preprint arXiv:2303.18223,

    2023 arXiv

  29. [44]

    Dynllm: When large language models meet dynamic graph recommendation

    [Zhao et al., 2024b] Ziwei Zhao, Fake Lin, Xi Zhu, et al. Dynllm: When large language models meet dynamic graph recommendation. arXiv preprint arXiv:2405.07580, 2024

    2024 arXiv

  30. [2016]

    Comprehending knowledge graphs with large lan- guage models for recommender systems

    [Cui et al., 2024] Ziqiang Cui, Yunpeng Weng, Xing Tang, et al. Comprehending knowledge graphs with large lan- guage models for recommender systems. arXiv preprint arXiv:2410.12229,

    2024 arXiv

  31. [2019]

    A survey of graph meets large language model: Progress and future directions

    [Li et al., 2023] Yuhan Li, Zhixun Li, Peisong Wang, et al. A survey of graph meets large language model: Progress and future directions. arXiv preprint arXiv:2311.12399,

    2023 arXiv

  32. [2020]

    Graph learning based recommender systems: A review

    [Wang et al., 2021] S Wang, L Hu, Y Wang, et al. Graph learning based recommender systems: A review. In Proc. of IJCAI. International Joint Conferences on Artificial In- telligence Organization,

    2021

  33. [2021]

    A prompting-based representation learning method for recommendation with large language models

    [Chen and Suzumura, 2024] Junyi Chen and Toyotaro Suzu- mura. A prompting-based representation learning method for recommendation with large language models. arXiv preprint arXiv:2409.16674,

    2024 arXiv

  34. [2023]

    Enhancing collaborative semantics of language model- driven recommendations via graph-aware learning

    [Guan et al., 2024] Zhong Guan, Likang Wu, Hongke Zhao, et al. Enhancing collaborative semantics of language model- driven recommendations via graph-aware learning. arXiv preprint arXiv:2406.13235,

    2024 arXiv

  35. [2024]

    On the opportunities and risks of foundation models

    [Bommasani et al., 2021] Rishi Bommasani, Drew A Hud- son, Ehsan Adeli, et al. On the opportunities and risks of foundation models. arXiv preprint arXiv:2108.07258,

    2021 arXiv

  36. [2025]

    Large language models on graphs: A comprehensive survey.IEEE Transactions on Knowledge and Data Engineering ,

    [Jin et al., 2024] Bowen Jin, Gang Liu, Chi Han, et al. Large language models on graphs: A comprehensive survey.IEEE Transactions on Knowledge and Data Engineering ,

    2024

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.