Pith. sign in

REVIEW 2 major objections 5 minor 40 references

Graph neural network inside LoRA boosts LLM recommendations

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · glm-5.2

2026-07-05 15:05 UTC pith:V3Z26PHH

load-bearing objection GNN inside the LoRA bottleneck is a clean new design point; gains over CoRA are real but small and lack significance testing the 2 major comments →

arxiv 2606.07526 v1 pith:V3Z26PHH submitted 2026-04-20 cs.CL cs.AI

GraphLoRA: Structure-Aware Low-Rank Adaptation for Large Language Model Recommendation

Lin Mu , Guoji Wang , Li Ni , Lei Sang , Zhize Wu , Peiquan Jin , Yiwen Zhang This is my paper
classification cs.CL cs.AI
keywords LoRAgraph neural networksrecommender systemslarge language modelslow-rank adaptationcollaborative filteringparameter-efficient fine-tuningstructure-aware alignment
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

This paper claims that the standard way of adapting large language models for recommendation—injecting collaborative signals as static embeddings—fails to capture the relational structure of user–item interaction graphs. The authors propose GraphLoRA, which inserts a trainable graph message-passing module directly into the low-rank adaptation bottleneck, between the down-projection and up-projection matrices. The central mechanism is that graph-structured neighborhood information is compressed into the same low-rank latent space where LoRA operates, fused with the semantic signal at that bottleneck, and then projected back to update the LLM's parameters. This converts LoRA from a passive fine-tuning adapter into a structure-aware reasoning pathway, where collaborative topology explicitly guides parameter updates during training. The paper reports that this design outperforms prior methods (including CoRA, CoLLM, and BinLLM) on ML-1M and Amazon-Book while using a stricter parameter budget (rank 8 on two projection types, single-layer injection), and that the advantage is most pronounced in cold-start scenarios where neighborhood aggregation compensates for sparse user histories.

Core claim

The paper's central discovery is that placing a graph neural network between the two low-rank matrices of LoRA (the down-projection A and up-projection B) produces better recommendation accuracy than injecting pre-computed collaborative embeddings, because the graph topology is dynamically aggregated and jointly optimized with the LLM rather than treated as a static input. The fusion happens in the compressed rank-r space, keeping parameter overhead near 1.7% above plain LoRA while the GNN's gradients flow back through the up-projection, ensuring the structural signal becomes semantically aligned with the LLM's reasoning. Under matched parameter budgets, the prior best method (CoRA-MF) degrt

What carries the argument

GraphLoRA inserts a message-passing GNN between LoRA's down-projection A and up-projection B, fusing graph-aggregated neighborhood signals with semantic features in the low-rank bottleneck before projecting back to the LLM's parameter space.

Load-bearing premise

The method's reported numbers depend on selecting which LLM layer receives the graph injection via validation tuning, and the paper shows roughly a 6-point AUC swing across layers. If the optimal layer cannot be determined without labeled validation data for each new dataset, the method's practical generality is limited by that hyperparameter sensitivity.

What would settle it

If the performance gain over CoRA vanishes under matched parameter budgets and matched injection layers, or if the optimal injection layer is shown to be random rather than systematically deep, the central architectural claim would be undermined.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • If structure-aware LoRA bottlenecks are effective for recommendation graphs, the same pattern—embedding a domain-specific relational encoder inside the low-rank adaptation pathway—could apply to other structured domains like knowledge graphs, molecular graphs, or social networks.
  • The finding that single-layer injection at deep layers outperforms multi-layer injection suggests that LLMs may have specific 'structural reasoning' layers where external graph signals are most compatible, which could guide where to place other types of auxiliary modules.
  • The cold-start advantage reported by the paper implies that dynamically aggregated neighborhood context partially substitutes for missing interaction history, which matters for new-user onboarding in production systems.

Where Pith is reading between the lines

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

  • The paper tunes the injection layer (layer 31 for ML-1M, layer 15 for Amazon-Book) and balancing coefficients via validation. If the optimal layer is dataset-specific, a principled selection criterion would be needed for deployment; one might hypothesize that the optimal depth correlates with the layer at which the LLM transitions from syntactic to semantic processing, which could be probed via re
  • The 1-hop neighbor sampling strategy limits the GNN's receptive field. Extending to multi-hop aggregation might capture higher-order collaborative signals, but the trade-off between receptive field and computational cost in the low-rank bottleneck is unexplored.
  • The fusion formula is a simple weighted sum of semantic and structural signals. A multiplicative or attention-based fusion mechanism might better capture interactions between the two signal types, especially for tokens that carry both textual and collaborative meaning.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 5 minor

Summary. The paper proposes GraphLoRA, which inserts a trainable graph message-passing module between the LoRA down-projection A and up-projection B, enabling structure-aware low-rank adaptation for LLM-based recommendation. The method initializes learnable MF embeddings, encodes them via a GNN, projects the structural signal through a bottleneck into the LoRA latent space, and injects it at a single Transformer layer. Experiments on ML-1M and Amazon-Book show improvements over CoRA and other baselines under AUC, UAUC, and NDCG@10. Code is publicly available.

Significance. The core architectural idea—placing a GNN inside the LoRA bottleneck rather than in the input space—is a clean and novel contribution to the LLM-rec alignment literature. The parameter-efficiency analysis (Table 6: 1.017x overhead vs. 7.635x for CoRA-MF) and the injection-position ablation (Pre-A vs. Middle) are well-motivated and falsifiable. The matched-budget comparison (Table 4) and the graph-encoder variant ablation (Table 5B) are appropriate experimental choices. The public code release is a notable strength for reproducibility.

major comments (2)
  1. §5.2, Table 3: The central claim of 'consistent' outperformance rests on single-run results with no error bars, standard deviations, or significance tests. The margins over the strongest baseline (CoRA-MF) are 1.1 AUC points on ML-1M (0.7472 vs. 0.7361) and 0.26 AUC points on Amazon-Book (0.8205 vs. 0.8179). In collaborative filtering benchmarks, AUC differences under 1 point are frequently within run-to-run noise. Without repeated runs (e.g., 3-5 seeds) and significance tests, the claim of consistent superiority cannot be distinguished from variance. This is load-bearing for the paper's central thesis.
  2. §5.2, Table 4: The matched-budget comparison shows CoRA-MF collapsing to UAUC 0.4995 on Amazon-Book under r=8/{q,v}, which is near-random. It is unclear whether CoRA-MF was re-tuned (learning rate, etc.) for this constrained setting or whether the r=16 hyperparameters were directly applied at r=8. If the latter, the large gap may reflect hyperparameter mismatch rather than architectural superiority. The paper should clarify the re-tuning protocol for the constrained baseline.
minor comments (5)
  1. §4.4, Eq. (9): The balancing coefficients λ_lora=1.0 and λ_gnn=0.1 are stated without justification. A brief sensitivity analysis or rationale for this ratio would strengthen the presentation.
  2. §5.1: The injection layer (layer 31 for ML-1M, layer 15 for Amazon-Book) is selected via validation tuning, and Figure 4 shows AUC varies by ~6 points across layers. A principled criterion or at least a discussion of why deeper layers are better for one dataset but not the other would improve practical guidance.
  3. Table 3: The 'second-best underlined' formatting is not visible in the text rendering; please ensure formatting is correct in the final version.
  4. §4.3, Eq. (6): The GNN update uses generic ψ and AGG functions but §5.1 specifies NGCF. Stating the concrete instantiation earlier (or in the equation caption) would reduce ambiguity.
  5. Figure 3: The warm/cold-start split criterion (interaction frequency threshold) is not specified. Please state the threshold used.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful and constructive review. The referee raises two major comments, both concerning experimental rigor: (1) the absence of repeated runs and significance tests in Table 3, and (2) potential hyperparameter mismatch in the matched-budget comparison (Table 4). Both points are well-taken. We will address them in the revised manuscript as follows. For Comment 1, we will re-run all main experiments (Table 3) with 5 random seeds and report mean ± standard deviation, along with paired t-tests against the strongest baseline (CoRA-MF). For Comment 2, we will clarify the re-tuning protocol for the constrained CoRA-MF baseline and, importantly, add a fully re-tuned CoRA-MF variant to eliminate the possibility that the collapse is an artifact of hyperparameter mismatch. We agree that these additions are necessary to substantiate the central claim of consistent outperformance.

read point-by-point responses

  1. Referee: §5.2, Table 3: The central claim of 'consistent' outperformance rests on single-run results with no error bars, standard deviations, or significance tests. The margins over the strongest baseline (CoRA-MF) are 1.1 AUC points on ML-1M (0.7472 vs. 0.7361) and 0.26 AUC points on Amazon-Book (0.8205 vs. 0.8179). In collaborative filtering benchmarks, AUC differences under 1 point are frequently within run-to-run noise. Without repeated runs (e.g., 3-5 seeds) and significance tests, the claim of consistent superiority cannot be distinguished from variance. This is load-bearing for the paper's central thesis.

    Authors: The referee is correct that single-run results without significance tests are insufficient to support a claim of consistent outperformance, particularly when the margins are within the range commonly observed as run-to-run noise in collaborative filtering. We accept this criticism. In the revised manuscript, we will re-run all experiments in Table 3 with 5 random seeds and report mean ± standard deviation for all methods. We will additionally conduct paired t-tests (or Wilcoxon signed-rank tests where appropriate) between GraphLoRA and the strongest baseline (CoRA-MF) on both datasets and all metrics. We will temper the language of 'consistent outperformance' to reflect only differences that are statistically significant. If any comparison fails to reach significance, we will state this transparently rather than claiming superiority. revision: yes

  2. Referee: §5.2, Table 4: The matched-budget comparison shows CoRA-MF collapsing to UAUC 0.4995 on Amazon-Book under r=8/{q,v}, which is near-random. It is unclear whether CoRA-MF was re-tuned (learning rate, etc.) for this constrained setting or whether the r=16 hyperparameters were directly applied at r=8. If the latter, the large gap may reflect hyperparameter mismatch rather than architectural superiority. The paper should clarify the re-tuning protocol for the constrained baseline.

    Authors: The referee raises a valid concern. We clarify that in the current version, CoRA-MF under the constrained setting (r=8, {q,v}) used the same hyperparameters as the r=16 configuration, with only a learning rate sweep over {5e-4, 1e-4, 5e-5}. We agree that this protocol may not constitute a fully fair re-tuning and that the near-random UAUC could partly reflect hyperparameter mismatch rather than a fundamental architectural limitation. In the revision, we will (1) explicitly state the re-tuning protocol used, acknowledging its limitations, and (2) conduct a more thorough hyperparameter search for the constrained CoRA-MF baseline, including learning rate, LoRA scaling factor alpha, weight decay, and warmup schedule. We will report the best result obtained under this expanded search. If the collapse persists after thorough re-tuning, this strengthens our architectural argument; if it does not, we will report the improved baseline honestly and reframe the contribution accordingly. Either way, the revised manuscript will present a fairer comparison. revision: yes

Circularity Check

0 steps flagged

No circularity found — the derivation chain is self-contained with external graph data as input

full rationale

The paper's derivation chain is: (1) learnable MF embeddings E=[P;Q] are initialized from user-item interaction data (§4.1, Eq. 4); (2) graph structural encoding performs message passing on the user-item bipartite graph G=(U,I,E) using E as node features (§4.3, Eq. 6), producing structure-enhanced representations e_n^(L); (3) these are projected through a bottleneck W_neck into the LoRA rank-r space (Eq. 7); (4) the structural signal z_n is fused with the semantic signal h_sem=Ax_t via h_latent = λ_lora·h_sem + λ_gnn·z_n (Eq. 9), then projected back by B (Eq. 10). At no point does an output reduce to its own input by construction. The graph topology is external data (user-item interactions), not derived from the LLM or the target metric. The MF embeddings E are jointly optimized end-to-end, which is standard training, not circular reasoning. The hyperparameters λ_lora=1.0, λ_gnn=0.1 and injection layer l* are tuned on validation data (§5.1), which is standard practice and does not constitute fitted-input-as-prediction. The one self-citation (DenseLoRA, Mu et al. 2025) appears only in related works (§2.2) and is not load-bearing for any architectural or empirical claim. The concerns about missing error bars and potential baseline under-tuning are correctness/statistical issues, not circularity. The derivation is self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

6 free parameters · 4 axioms · 0 invented entities

The paper introduces no new postulated entities (no new particles, forces, dimensions, or conserved quantities). All components — LoRA, GNN, MF embeddings, projection layers — are standard. The contribution is the specific architectural combination. The free parameters are all hyperparameters tuned on validation data, which is standard ML practice but should be noted.

free parameters (6)
  • λ_lora = 1.0
    Balancing coefficient for the semantic signal in the structural fusion step (Eq. 9). Set by validation tuning per §5.1.
  • λ_gnn = 0.1
    Balancing coefficient for the graph structural signal in the fusion step (Eq. 9). Set by validation tuning per §5.1.
  • Injection layer l* = 31 (ML-1M), 15 (Amazon-Book)
    The specific Transformer layer at which the GNN signal is injected. Selected via validation tuning (§5.1, Figure 4).
  • MF embedding dimension d_emb = 256
    Dimensionality of the collaborative embedding matrix E. Stated in §5.1.
  • LoRA rank r = 8
    Intrinsic rank of the low-rank adaptation. Stated in §5.1.
  • GNN depth L = 3
    Number of message-passing layers in the NGCF graph encoder. Stated in §5.1.
axioms (4)
  • domain assumption LoRA's low-rank bottleneck (the r-dimensional latent space between A and B) is a suitable subspace for fusing graph-structured signals with semantic signals.
    The entire architecture rests on the assumption that injecting graph signals into the r-dimensional LoRA bottleneck is effective. This is not proven from first principles; it is validated empirically through ablation (Table 5) and the matched-budget comparison (Table 4).
  • domain assumption A single-layer injection strategy is sufficient to capture the benefits of structure-aware propagation.
    §4.4 adopts a sparse single-layer injection at layer l*. The paper does not test multi-layer injection, assuming that one layer is enough. The layer-sweep (Figure 4) shows sensitivity but only varies the single injection point.
  • domain assumption Matrix factorization embeddings provide a valid initialization for collaborative identity that can be jointly refined with the LLM.
    §4.1 uses MF embeddings E as the shared learnable state. This follows CoRA's precedent but assumes MF is an adequate starting representation for the graph encoder.
  • standard math Standard message-passing GNNs (NGCF, LightGCN, GCN) can effectively aggregate collaborative signals on sampled subgraphs for injection.
    §3.2 and §4.3 use the MPNN paradigm. This is a well-established framework (Gilmer et al., 2017).

pith-pipeline@v1.1.0-glm · 16938 in / 3214 out tokens · 185762 ms · 2026-07-05T15:05:58.948454+00:00 · methodology

0 comments
read the original abstract

Large Language Models (LLMs) have shown strong potential for recommendation (LLMRec) due to their powerful reasoning and generalization abilities. However, effectively aligning the textual semantics modeled by LLMs with the collaborative signals remains a key challenge. Existing methods either translate collaborative information into textual prompts or inject pre-trained embeddings into the LLM, both of which treat structural information as static input and fail to capture high-order relational dependencies. To bridge this gap, we propose GraphLoRA, a novel framework that generalizes low-rank adaptation from independent to structure-aware propagation. GraphLoRA embeds a trainable graph message-passing network within the low-rank adaptation pathway, enabling structural signals to propagate through the parameter space. This design allows collaborative topology to explicitly guide parameter updates, fostering deep integration between graph-structured and textual semantic information. Extensive experiments on multiple benchmarks demonstrate that GraphLoRA not only outperforms state-of-the-art LLM-based recommendation methods but also achieves superior generalization, effectively balancing structural reasoning capability with computational efficiency. Code is available at \href{https://github.com/wgj15965/GraphLoRA}{https://github.com/wgj15965/GraphLoRA}.

Figures

Figures reproduced from arXiv: 2606.07526 by Guoji Wang, Lei Sang, Li Ni, Lin Mu, Peiquan Jin, Yiwen Zhang, Zhize Wu.

Figure 1
Figure 1. Figure 1: Comparison of collaborative alignment paradigms. (a) Input-Space Alignment converts in￾teraction histories into textual prompts for the LLM input. (b) Parameter-Space Alignment injects static, externally encoded embeddings into LLM weights (e.g., via LoRA). (c) Structure-Aware Alignment (Ours) integrates a learnable GNN within the LoRA bottleneck (between A and B) to perform dynamic message pass￾ing direct… view at source ↗
Figure 2
Figure 2. Figure 2: Overview of GraphLoRA. (Left) The end-to-end recommendation process fusing text prompts and collaborative signals. (Right) The detailed structure of the GraphLoRA layer, where a GNN-based structural encoder is injected between the low-rank matrices A and B to realize structure-aware low-rank propagation. Note: “Embs” refers to “Embeddings”. collaborative tokens (i.e., users xuserid and items xitemid). Here… view at source ↗
Figure 3
Figure 3. Figure 3: Warm/cold evaluation on Amazon-Book and ML-1M. The left and right y-axes correspond to AUC and UAUC, respectively. tion with a small rank(e.g., r=8) at a single-layer. Despite this compact design, GraphLoRA achieves comparable or better performance, highlighting a more favorable performance–efficiency trade-off. Warm and Cold Start Analysis. We split the test set into warm and cold subsets based on user in… view at source ↗
Figure 4
Figure 4. Figure 4: Effect of the GNN insertion layer on ML-1M [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗

discussion (0)

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

Reference graph

Works this paper leans on

40 extracted references · 40 canonical work pages · 3 internal anchors

  1. [1]

    Keqin Bao, Jizhi Zhang, Yang Zhang, Wenjie Wang, Fuli Feng, and Xiangnan He. 2023. TALLRec : An effective and efficient tuning framework to align large language model with recommendation. In Proceedings of the 17th ACM Conference on Recommender Systems (RecSys), RecSys '23, pages 1007--1014. Association for Computing Machinery

    work page 2023

  2. [2]

    Gonzalez, Ion Stoica, and Eric P

    Wei-Lin Chiang, Zhuohan Li, Zi Lin, Ying Sheng, Zhanghao Wu, Hao Zhang, Lianmin Zheng, Siyuan Zhuang, Yonghao Zhuang, Joseph E. Gonzalez, Ion Stoica, and Eric P. Xing. 2023. Vicuna: An open-source chatbot impressing GPT -4 with 90\

    work page 2023

  3. [3]

    Sunhao Dai, Ninglu Shao, Haiyuan Zhao, Weijie Yu, Zihua Si, Chen Xu, Zhongxiang Sun, Xiao Zhang, and Jun Xu. 2023. Uncovering chatgpt’s capabilities in recommender systems. In Proceedings of the 17th ACM Conference on Recommender Systems (RecSys), pages 1126--1132

    work page 2023

  4. [4]

    Ning Ding, Yujia Qin, Guang Yang, Fuchao Wei, Zonghan Yang, Yusheng Su, Shengding Hu, Yulin Chen, Chi-Min Chan, Weize Chen, et al. 2023. Parameter-efficient fine-tuning of large-scale pre-trained language models. Nature Machine Intelligence, 5(3):220--235

    work page 2023

  5. [5]

    Qingxiu Dong, Lei Li, Damai Dai, Ce Zheng, Jingyuan Ma, Rui Li, Heming Xia, Jingjing Xu, Zhiyong Wu, Tianyu Liu, Baobao Chang, Xu Sun, Lei Li, and Zhifang Sui. 2024. https://arxiv.org/abs/2301.00234 A survey on in-context learning . Preprint, arXiv:2301.00234

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  6. [6]

    Yunfan Gao, Tao Sheng, Youlin Xiang, Yun Xiong, Haofen Wang, and Jiawei Zhang. 2023. Chat-rec: Towards interactive and explainable llms-augmented recommender system. arXiv preprint arXiv:2303.14524

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  7. [7]

    Shijie Geng, Shuchang Liu, Zuohui Fu, Yingqiang Ge, and Yongfeng Zhang. 2022. Recommendation as language processing (rlp): A unified pretrain, personalized prompt & predict paradigm (p5). In Proceedings of the 16th ACM Conference on Recommender Systems (RecSys), pages 299--315

    work page 2022

  8. [8]

    Schoenholz, Patrick F

    Justin Gilmer, Samuel S. Schoenholz, Patrick F. Riley, Oriol Vinyals, and George E. Dahl. 2017. Neural message passing for quantum chemistry. In Proceedings of the 34th International Conference on Machine Learning (ICML)

    work page 2017

  9. [9]

    Will Hamilton, Zhitao Ying, and Jure Leskovec. 2017. Inductive representation learning on large graphs. In Advances in Neural Information Processing Systems (NeurIPS), volume 30

    work page 2017

  10. [10]

    Maxwell Harper and Joseph A

    F. Maxwell Harper and Joseph A. Konstan. 2015. The movielens datasets: History and context. ACM Transactions on Interactive Intelligent Systems, 5(4):1--19

    work page 2015

  11. [11]

    Ruining He and Julian McAuley. 2016. Ups and downs: Modeling the visual evolution of fashion trends with one-class collaborative filtering. In Proceedings of the 25th International Conference on World Wide Web, pages 507--517

    work page 2016

  12. [12]

    Xiangnan He, Kuan Deng, Xiang Wang, Yan Li, Yongdong Zhang, and Meng Wang. 2020. LightGCN : Simplifying and powering graph convolution network for recommendation. In Proceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR), pages 639--648

    work page 2020

  13. [13]

    Neil Houlsby, Andrei Giurgiu, Stanislaw Jastrzebski, Bruna Morrone, Quentin De Laroussilhe, Andrea Gesmundo, Mona Attariyan, and Sylvain Gelly. 2019. Parameter-efficient transfer learning for nlp. In International Conference on Machine Learning (ICML), pages 2790--2799. PMLR

    work page 2019

  14. [14]

    Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen

    Edward J. Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. 2022. LoRA : Low-rank adaptation of large language models. In International Conference on Learning Representations (ICLR)

    work page 2022

  15. [15]

    Lei Huang, Weijiang Yu, Weitao Ma, Weihong Zhong, Zhangyin Feng, Haotian Wang, Qianglong Chen, Weihua Peng, Xiaocheng Feng, Bing Qin, and Ting Liu. 2024. A survey on hallucination in large language models: Principles, taxonomy, challenges, and open questions. ACM Transactions on Information Systems

    work page 2024

  16. [16]

    Mingyu Jin, Qinkai Yu, Jingyuan Huang, Qingcheng Zeng, Zhenting Wang, Wenyue Hua, Haiyan Zhao, Kai Mei, Yanda Meng, Kaize Ding, Fan Yang, Mengnan Du, and Yongfeng Zhang. 2025. Exploring concept depth: How large language models acquire knowledge and concept at different layers? In Proceedings of the 31st International Conference on Computational Linguistic...

    work page 2025

  17. [17]

    Wang-Cheng Kang and Julian McAuley. 2018. Self-attentive sequential recommendation. In 2018 IEEE International Conference on Data Mining (ICDM), pages 197--206. IEEE

    work page 2018

  18. [18]

    Kipf and Max Welling

    Thomas N. Kipf and Max Welling. 2017. Semi-supervised classification with graph convolutional networks. In International Conference on Learning Representations (ICLR)

    work page 2017

  19. [19]

    Yehuda Koren, Robert Bell, and Chris Volinsky. 2009. Matrix factorization techniques for recommender systems. Computer, 42(8):30--37

    work page 2009

  20. [20]

    Lei Li, Yongfeng Zhang, and Li Chen. 2023. Personalized prompt learning for explainable recommendation. ACM Transactions on Information Systems, 41(4):1--26

    work page 2023

  21. [21]

    Xiang Lisa Li and Percy Liang. 2021. Prefix-tuning: Optimizing continuous prompts for generation. In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics (ACL), pages 4582--4597

    work page 2021

  22. [22]

    Jiayi Liao, Sihang Li, Zhengyi Yang, Jiancan Wu, Yancheng Yuan, Xiang Wang, and Xiangnan He. 2024. LLaRA : Large language-recommendation assistant. In Proceedings of the 47th International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR), pages 1785--1795

    work page 2024

  23. [23]

    Qidong Liu, Xian Wu, Wanyu Wang, Yejing Wang, Yuanshao Zhu, Xiangyu Zhao, Feng Tian, and Yefeng Zheng. 2025 a . LLMEmb : Large language model can be a good embedding generator for sequential recommendation. In Proceedings of the AAAI Conference on Artificial Intelligence (AAAI), volume 39, pages 12183--12191

    work page 2025

  24. [24]

    Yiyu Liu, Qian Liu, Yu Tian, Changping Wang, Yanan Niu, Yang Song, and Chenliang Li. 2021. Concept-aware denoising graph neural network for micro-video recommendation. In Proceedings of the 30th ACM International Conference on Information and Knowledge Management, pages 1099--1108

    work page 2021

  25. [25]

    Yuting Liu, Jinghao Zhang, Yizhou Dang, Yuliang Liang, Qiang Liu, Guibing Guo, Jianzhe Zhao, and Xingwei Wang. 2025 b . Cora: Collaborative information perception by large language model's weights for recommendation. In Proceedings of the AAAI Conference on Artificial Intelligence (AAAI), volume 39, pages 12246--12254

    work page 2025

  26. [26]

    Lin Mu, Xiaoyu Wang, Li Ni, Yang Li, Zhize Wu, Peiquan Jin, and Yiwen Zhang. 2025. D ense L o RA : Dense low-rank adaptation of large language models. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics, pages 10198--10211. Association for Computational Linguistics

    work page 2025

  27. [27]

    Oscar Skean, Md Rifat Arefin, Yann LeCun, and Ravid Shwartz-Ziv. 2024. Does representation matter? exploring intermediate layers in large language models. arXiv preprint arXiv:2412.09563

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  28. [28]

    Jiabin Tang, Yuhao Yang, Wei Wei, Lei Shi, Lixin Su, Suqi Cheng, Dawei Yin, and Chao Huang. 2024 a . GraphGPT : Graph instruction tuning for large language models. In Proceedings of the 47th International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR), pages 491--500

    work page 2024

  29. [29]

    Jiabin Tang, Yuhao Yang, Wei Wei, Lei Shi, Long Xia, Dawei Yin, and Chao Huang. 2024 b . HiGPT : Heterogeneous graph language model. In Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD), pages 2842--2853

    work page 2024

  30. [30]

    Petar Veli c kovi \'c , Guillem Cucurull, Arantxa Casanova, Adriana Romero, Pietro Li \`o , and Yoshua Bengio. 2018. Graph attention networks. In International Conference on Learning Representations (ICLR)

    work page 2018

  31. [31]

    Xiang Wang, Xiangnan He, Meng Wang, Fuli Feng, and Tat-Seng Chua. 2019. Neural graph collaborative filtering. In Proceedings of the 42nd International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR), pages 165--174

    work page 2019

  32. [32]

    Likang Wu, Zhi Zheng, Zhaopeng Qiu, Hao Wang, Hongchao Gu, Tingjia Shen, Chuan Qin, Chen Zhu, Hengshu Zhu, Qi Liu, et al. 2024. A survey on large language models for recommendation. World Wide Web, 27(5):60

    work page 2024

  33. [33]

    Shiwen Wu, Fei Sun, Wentao Zhang, Xu Xie, and Bin Cui. 2022. Graph neural networks in recommender systems: A survey. ACM Computing Surveys, 55(5):1--37

    work page 2022

  34. [34]

    Yang Zhang, Keqin Bao, Ming Yan, Wenjie Wang, Fuli Feng, and Xiangnan He. 2024. Text-like encoding of collaborative information in large language models for recommendation. In Proceedings of the Annual Meeting of the Association for Computational Linguistics (ACL), pages 9061--9079

    work page 2024

  35. [35]

    Yang Zhang, Fuli Feng, Jizhi Zhang, Keqin Bao, Qifan Wang, and Xiangnan He. 2025. Collm: Integrating collaborative embeddings into large language models for recommendation. IEEE Transactions on Knowledge and Data Engineering, 37(5):2329--2340

    work page 2025

  36. [36]

    Zizhuo Zhang and Bang Wang. 2023. Prompt learning for news recommendation. In Proceedings of the 46th International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR), pages 1157--1166

    work page 2023

  37. [37]

    Haiteng Zhao, Shengchao Liu, Chang Ma, Hannan Xu, Jie Fu, Zhi-Hong Deng, Lingpeng Kong, and Qi Liu. 2023. Gimlet: A unified graph-text model for instruction-based molecule zero-shot learning. In Advances in Neural Information Processing Systems (NeurIPS), volume 36, pages 6485--6509

    work page 2023

  38. [38]

    Cai-Nicolas Ziegler. 2005. Improving recommendation lists through topic diversification. In WWW'05: Proceedings of the 14th international conference on World Wide Web, pages 22--32. ACM

    work page 2005

  39. [39]

    online" 'onlinestring :=

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint eprinttype howpublished institution journal key month note number organization pages publisher school series title type volume year doi pubmed url lastchecked label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block STRING...

    work page

  40. [40]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

    work page