Pith. sign in

REVIEW 2 major objections 5 minor 76 references

The paper claims that in LLM-based next-item ranking, knowledge graphs should constrain the ranking decision itself—through representation, context, candidate comparison, and score calibration—not just enrich the prompt.

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 · deepseek-v4-flash

2026-08-04 00:51 UTC pith:FOYWVX2C

load-bearing objection A well-specified decision-level graph-grounding architecture for LLM recommendation, but the headline result is likely inflated by an explicit popularity prior that the random-negative protocol rewards—recheck before trusting the numbers. the 2 major comments →

arxiv 2608.00669 v1 pith:FOYWVX2C submitted 2026-08-01 cs.IR

GARDRec: Decision-Level Graph Grounding for Large Language Model Recommendation

Yong Wang , Hongliang Sun , Jinlan Liu , Hua Zhang , Dianbo Sui , Dianhui Chu , Zhiying Tu This is my paper
classification cs.IR
keywords Large language modelsRecommendationKnowledge graphGraph-grounded recommendationNext-item rankingRetrieval-augmented generationDecision-level groundingReranking
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.

The paper argues that in LLM-based next-item recommendation, a knowledge graph should participate in the ranking decision itself—through item representation, context construction, candidate comparison, and score calibration—rather than only being serialized into the prompt. It introduces GARDRec, which turns graph embeddings into continuous prompts aligned with a frozen LLM, then adds late-stage scoring branches for interaction and matching features that bypass the text prompt. Across three datasets and three LLM backbones, GARDRec reports the best result in 24 of 27 metric columns compared with prompt-level and graph-token baselines. The authors frame the finding as a design principle: graph evidence is most useful when it constrains the final decision.

Core claim

The central claim is that graph evidence is more effective when it participates in representation learning, context construction, candidate comparison, and score calibration, rather than only enriching the prompt. GARDRec embodies this by learning semantic-structural item embeddings from textual node features and graph propagation; constructing a personalized graph context from a recency-weighted centroid of historical items and their first-order neighborhoods; aligning these graph-derived vectors with the hidden space of a frozen LLM through a projector and soft markers; and scoring candidates with an inter-candidate attention layer plus explicit interaction and hard-matching features that

What carries the argument

The load-bearing mechanism is the split of information channels. Graph structure and semantics enter the LLM as continuous multimodal prompts—projected graph embeddings with position and slot embeddings, separated by soft markers—so the LLM can reason over them. Precise numerical signals (graph similarity, transition statistics, co-occurrence, popularity, attribute overlap, retrieval rank) are routed to late-stage scoring branches outside the text sequence, avoiding dilution by attention over long prompts. Inter-candidate attention then lets each candidate be scored against its competitors, and a final score sums the discriminative logit with a log-normalized generative likelihood restricted

Load-bearing premise

The empirical support rests on a leave-one-out protocol where each test instance contains one true item and 19 randomly sampled unobserved candidates; if that pool does not reflect real ranking difficulty, the reported 24-of-27 advantage may shrink.

What would settle it

Run the same GARDRec versus prompt-level KG-RAG comparison with the same models but replace the 19 random negatives with 19 hard negatives—items the user almost interacted with or top-ranked popular items—and expand the pool to several hundred candidates. If the advantage over prompt-level KG-RAG disappears or reverses, the claim that decision-level grounding improves ranking is not general.

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

If this is right

  • Prompt-level KG-RAG is a weak constraint for next-item ranking: merely providing graph facts as text does not guarantee they shape the final choice.
  • Graph representations aligned with the LLM hidden space can sharply improve top-1 accuracy over text-only retrieval (ACC from 0.128 to 0.444 in the Qwen2-7B/ML-1M diagnostic).
  • Explicit ranking supervision, inter-candidate attention, and generative calibration each contribute complementary gains, so each is worth keeping in graph-grounded LLM recommenders.
  • The advantage is strongest for users with 1–3 history items, making decision-level grounding a candidate solution for sparse-history ranking.
  • For pure cold-start users with no history, an item-only KG prompt fallback is more effective, defining the operating boundary of the method.

Where Pith is reading between the lines

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

  • If the principle generalizes, other LLM ranking tasks (search reranking, QA answer selection, tool selection) may also benefit from routing fine-grained numerical features to decision layers instead of the prompt.
  • The random-negative 20-candidate protocol likely understates the difficulty of real top-N ranking; a testable extension is to evaluate under hard-negative or full-catalog pools, where inter-candidate attention and explicit features may matter even more—or less.
  • The framework suggests a hybrid deployment: use prompt-level KG verbalization for zero-history users and GARDRec-style decision grounding once any interaction exists.
  • Since the generative branch only uses candidate labels, a next step is to test whether a learned answer head with item-specific decoding would further calibrate the ranking distribution.

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 GARDRec, a framework that uses knowledge graphs as a decision-level grounding layer for LLM-based next-item ranking, rather than only as prompt-level evidence. The method builds semantic-structural item embeddings through GraphSAGE over an item knowledge graph, constructs personalized graph contexts from temporally weighted user histories and first-order neighborhoods, aligns graph embeddings with a frozen LLM via a projection, and injects explicit interaction and matching features into late-stage scoring. It also introduces inter-candidate attention and a restricted generative likelihood. Experiments on ML-1M, ML-20M, and Amazon-Book with LLaMA2-7B, LLaMA3-8B, and Qwen2-7B report that GARDRec achieves the best result in 24 of 27 metric columns. Ablations and diagnostics on ML-1M/Qwen2-7B attribute gains to graph projection, neighborhood retrieval, late fusion, ranking loss, and generative calibration, and disclose a zero-history setting where the prompt-level baseline K-RagRec outperforms GARDRec.

Significance. If the empirical results are reliable, the paper makes a useful conceptual contribution by separating graph evidence into representation, context construction, candidate comparison, and score calibration. The framework is non-trivial, the evaluation covers multiple datasets and backbone LLMs, and the honest disclosure of the zero-history boundary condition is a strength. The limitations—single-run reporting and an explicit popularity channel—are addressable and do not invalidate the framework, but they currently prevent a strong acceptance.

major comments (2)
  1. [§4.1.2, Tables 2–6] All reported metrics are single-run point estimates; no standard deviations, number of seeds, or significance tests are provided. The headline claim of winning 24 of 27 metric columns in §4.2 could be within run-to-run noise, and several differences are very small (e.g., LLaMA3/ML-1M ACC 0.472 vs 0.471). Request mean±std over at least three seeds and paired significance tests over test users for the main comparison and the diagnostic analyses.
  2. [Eq. (6), Eq. (14), §4.1.2] Stats(u,v_t) in Eq. (6) includes item popularity, and Eq. (14) injects this feature into the ranking logit through f_interact. Under the M=20 protocol with one real next item and 19 random unobserved negatives, a popularity-only ranker can already place many targets at or near the top. No popularity-only baseline is reported, and the ablations remove entire branches (w/o Late Fusion, w/o Match Features) rather than isolating the Stats component. Please add a popularity-only baseline and an ablation that removes only Stats(u,v_t) while keeping the rest of f_interact, and report how much of the measured advantage survives. The tail-cold result in Table 5 is informative but does not fully resolve this confound.
minor comments (5)
  1. [§4.1.1] The text says ML-1M contains 'over 4,000 movies' but Table 1 reports 3,533 items; please reconcile.
  2. [Tables 4–5] Several numbers are run together without spaces, e.g., '8860.3815' and '4350.5425'. Fix formatting.
  3. [Eq. (14)] The fixed 0.3 coefficient on the matching branch appears without justification. State how it was chosen or make it learnable.
  4. [§4.5, abstract] The zero-history limitation is disclosed in §4.5, but the abstract and introduction state the claim for 'next-item ranking' without this operating condition. Please explicitly state that GARDRec targets sparse-but-non-empty user histories.
  5. [References] Several references are incomplete (e.g., Resnick et al., Koren et al., He et al. missing years). Please complete all bibliography entries.

Circularity Check

0 steps flagged

No significant circularity: the paper is an empirical system comparison whose predictions are trained on held-out targets and tested against external baselines, not derived from its own inputs by construction.

full rationale

GARDRec is a supervised ranking system evaluated under a leave-one-out protocol. The final score in Eq. (16) combines a discriminative logit, a late-fusion interaction/matching branch, and a restricted generative likelihood; it is optimized with cross-entropy, BPR, and LM losses (Eqs. 17-20) on training instances and then evaluated on held-out most-recent interactions. The paper explicitly states that 'All transition, co-occurrence, popularity, and retrieval-prior features are computed only from the training interactions and metadata, without using test targets' (Section 4.1.4), so the features entering Stats(u,v_t) and Rank(v_t) are not fit to the test labels. The central claim—that decision-level graph grounding outperforms prompt-level KG-RAG—is not definitional: it is tested against external baselines (KG-Text, KAPING, GraphToken, G-Retriever, K-RagRec) under the same candidate sets. The strict zero-history diagnostic (Table 6) reports a regime where K-RagRec beats GARDRec, which is not what a circular or tautological construction would produce. The skeptical concern about item-popularity priors and the M=20 random-negative protocol is a validity/confound question about the evaluation benchmark, not a circularity: a popularity-correlated feature is a model input whose contribution is learned and then tested on unseen labels, and no equation in the paper reduces the reported prediction to that feature by construction. Self-citation is not load-bearing: no centrality, uniqueness theorem, or ansatz is imported from prior work by the same authors; references such as K-RagRec and GraphToken are external baselines. Therefore no circular step meets the evidentiary bar.

Axiom & Free-Parameter Ledger

5 free parameters · 4 axioms · 0 invented entities

No new physical or theoretical entities are introduced; soft markers, slot embeddings, and answer-query tokens are learned parameter groups rather than independent postulated entities. The true ledger burden is the set of hand-set hyperparameters and the domain assumptions about KG quality and evaluation protocol.

free parameters (5)
  • Loss weights λ1 (BPR) and λ2 (LM) = λ1=0.3, λ2=0.1 (§4.1.4)
    Chosen by hand to balance three objectives; no search procedure reported.
  • Matching-branch coefficient 0.3 in Eq. 14 = 0.3
    Ad hoc weight controlling the hard-matching feature contribution to the ranking logit.
  • Generative calibration weight γ in Eq. 16 = Not reported
    Controls the generative branch in the final score; value and tuning method are not given.
  • Semantic-path balancing weights η1, η2 in Eq. 13 = Fixed, values not reported
    Weights for the recency and retrieval compatibility pathways in the semantic score.
  • History length N=10 and candidate count M=20 = N=10, M=20 (§4.1.2)
    Protocol constants; reported performance could depend on these choices.
axioms (4)
  • domain assumption The constructed item knowledge graph accurately links items to their attributes (TMDB for movies, product metadata for books).
    If item–entity alignment is noisy, graph propagation and neighborhood retrieval inject misleading context; Table 1 reports graph statistics but no quality check.
  • domain assumption First-order graph neighborhoods of historical items are a useful expansion of user preference.
    Eq. 5 averages first-order neighbors; if neighbors are uninformative, the retrieved context hurts rather than helps. The ablation shows a drop when removed, but only on one backbone-dataset setting.
  • domain assumption Leave-one-out candidate ranking with 20 random negatives faithfully measures next-item recommendation performance.
    §4.1.2 defines this protocol; random negatives and a single split differ from real production top-N ranking, and no repeated trials or significance tests are reported.
  • domain assumption A frozen LLM can be aligned with continuous projected graph embeddings by training only small adapter components.
    The multimodal prompt construction in §3.5 relies on the assumption that the LLM hidden space accepts non-text continuous tokens after a learned projection.

pith-pipeline@v1.3.0-daily-deepseek · 20692 in / 11966 out tokens · 116573 ms · 2026-08-04T00:51:58.945017+00:00 · methodology

0 comments
read the original abstract

Large language models (LLMs) offer new opportunities for recommendation by interpreting item descriptions, user instructions, and external knowledge through natural-language prompts. However, existing graph-augmented LLM recommenders often use knowledge graphs mainly as prompt-level evidence, leaving ranking decisions weakly constrained by structured user-item relations. This is problematic for next-item recommendation, where the model must compare candidates under the same user context while preserving temporal preference, collaborative signals, and attribute matches. To address this issue, we propose \emph{GARDRec}, a Graph-grounded Adaptive Reasoning and Decision-aware Recommendation framework for LLM-based next-item ranking. GARDRec constructs semantic-structural item representations from textual node features and graph propagation, derives personalized graph contexts from temporally weighted histories and first-order neighborhoods, and aligns graph-derived representations with a frozen LLM through continuous multimodal prompts. Explicit interaction and matching features are injected through late-stage decision branches, while inter-candidate attention and restricted generative likelihood support final ranking. Experiments on three public benchmarks with multiple LLM backbones show that GARDRec generally improves candidate-ranking performance over representative baselines. Ablation and diagnostic analyses verify the contributions of graph projection, neighborhood retrieval, explicit decision features, ranking loss, and generative calibration.

Figures

Figures reproduced from arXiv: 2608.00669 by Dianbo Sui, Dianhui Chu, Hongliang Sun, Hua Zhang, Jinlan Liu, Yong Wang, Zhiying Tu.

Figure 1
Figure 1. Figure 1: (a) Prompt-level graph augmentation exposes related facts to an LLM, but the final next-item decision can remain sensitive to order, context placement, and weakly modeled structured dependencies. (b) A decision-grounded graph recommender constructs personalized graph context from user history and knowledge-graph neighborhoods, allowing relational evidence to guide candidate comparison before final ranking.… view at source ↗
Figure 2
Figure 2. Figure 2: The overall framework of the proposed GARDRec model. The node representation is then updated by combining its previous state with the aggregated neighborhood representation: ℎ (𝑙) 𝑣 = 𝜎 ( 𝑊 (𝑙) [ ℎ (𝑙−1) 𝑣 ∥ ℎ (𝑙) (𝑣) ]) (3) where 𝑊 (𝑙) is the learnable transformation matrix, ∥ denotes vector concatenation, 𝜎(⋅) is a non-linear activation function, and AGGREGATE(𝑙) is a permutation-invariant mean aggregat… 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

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

  1. [1]

    Knowledge-

    Baek, Jinheon and Aji, Alham Fikri and Saffari, Amir , date =. Knowledge-. doi:10.48550/arXiv.2306.04136 , url =. 2306.04136 , eprinttype =

    Pith/arXiv arXiv doi:10.48550/arxiv.2306.04136

  2. [2]

    2024 , doi =

    Lin, Jianghao and Shan, Rong and Zhu, Chenxu and Du, Kounianhua and Chen, Bo and Quan, Shigang and Tang, Ruiming and Yu, Yong and Zhang, Weinan , booktitle =. 2024 , doi =

    2024

  3. [3]

    2024 , doi =

    Wu, Junda and Chang, Cheng-Chun and Yu, Tong and He, Zhankui and Wang, Jianing and Hou, Yupeng and McAuley, Julian , booktitle =. 2024 , doi =

    2024

  4. [4]

    2023 , doi =

    Zhang, Yang and Feng, Fuli and Zhang, Jizhi and Bao, Keqin and Wang, Qifan and He, Xiangnan , journal =. 2023 , doi =

    2023

  5. [5]

    2025 , doi =

    Meng, Zeyuan and Yi, Zixuan and Ounis, Iadh , journal =. 2025 , doi =

    2025

  6. [6]

    2025 , doi =

    Azizi, Vahid and Koochaki, Fatemeh , journal =. 2025 , doi =

    2025

  7. [7]

    2025 , doi =

    Kim, Sunwoo and Lee, Geon and Kim, Kyungho and Yoo, Jaemin and Shin, Kijung , journal =. 2025 , doi =

    2025

  8. [8]

    2025 , doi =

    Li, Yuhan and Zhang, Xinni and Luo, Linhao and Chang, Heng and Ren, Yuxiang and King, Irwin and Li, Jia , booktitle =. 2025 , doi =

    2025

  9. [9]

    Proceedings of the 17th

    Bao, Keqin and Zhang, Jizhi and Zhang, Yang and Wang, Wenjie and Feng, Fuli and He, Xiangnan , date =. Proceedings of the 17th. doi:10.1145/3604915.3608857 , url =

    arXiv

  10. [10]

    Unveiling and

    Das, Anindya Bijoy and Sakib, Shahnewaz Karim , date =. Unveiling and. doi:10.48550/arXiv.2409.10825 , url =. 2409.10825 , eprinttype =

    doi:10.48550/arxiv.2409.10825

  11. [11]

    Contextualizing Recommendation Explanations with LLMs: A User Study

    Feng, Yuanjun and Feuerriegel, Stefan and Shrestha, Yash Raj , date =. Contextualizing. doi:10.48550/arXiv.2501.12152 , url =. 2501.12152 , eprinttype =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2501.12152

  12. [12]

    Gao, Yunfan and Sheng, Tao and Xiang, Youlin and Xiong, Yun and Wang, Haofen and Zhang, Jiawei , date =. Chat-. doi:10.48550/arXiv.2303.14524 , url =. 2303.14524 , eprinttype =

    Pith/arXiv arXiv doi:10.48550/arxiv.2303.14524

  13. [13]

    doi:10.48550/arXiv.2406.12433 , url =

    Gao, Jingtong and Chen, Bo and Liu, Weiwen and Li, Xiangyang and Wang, Yichao and Wang, Wanyu and Guo, Huifeng and Tang, Ruiming and Zhao, Xiangyu , date =. doi:10.48550/arXiv.2406.12433 , url =. 2406.12433 , eprinttype =

    Pith/arXiv arXiv doi:10.48550/arxiv.2406.12433

  14. [14]

    and Laurent, Thomas and LeCun, Yann and Bresson, Xavier and Hooi, Bryan , date =

    He, Xiaoxin and Tian, Yijun and Sun, Yifei and Chawla, Nitesh V. and Laurent, Thomas and LeCun, Yann and Bresson, Xavier and Hooi, Bryan , date =. G-. doi:10.48550/arXiv.2402.07630 , url =. 2402.07630 , eprinttype =

    Pith/arXiv arXiv doi:10.48550/arxiv.2402.07630

  15. [15]

    doi:10.48550/arXiv.2002.02126 , url =

    He, Xiangnan and Deng, Kuan and Wang, Xiang and Li, Yan and Zhang, Yongdong and Wang, Meng , date =. doi:10.48550/arXiv.2002.02126 , url =. 2002.02126 , eprinttype =

    Pith/arXiv arXiv doi:10.48550/arxiv.2002.02126 2002

  16. [16]

    He, Xiangnan and Liao, Lizi and Zhang, Hanwang and Nie, Liqiang and Hu, Xia and Chua, Tat-Seng , date =. Neural. doi:10.48550/arXiv.1708.05031 , url =. 1708.05031 , eprinttype =

    Pith/arXiv arXiv doi:10.48550/arxiv.1708.05031

  17. [17]

    Hou, Yupeng and Zhang, Junjie and Lin, Zihan and Lu, Hongyu and Xie, Ruobing and McAuley, Julian and Zhao, Wayne Xin , date =. Large. doi:10.48550/arXiv.2305.08845 , url =. 2305.08845 , eprinttype =

    Pith/arXiv arXiv doi:10.48550/arxiv.2305.08845

  18. [18]

    Huang, Chao-Guang and Tian, Yu and Wu, Xiao-Ning and Xu, Zhan and Zhou, Bin , date =. New. doi:10.1088/0256-307X/29/4/040303 , url =. 0909.2773 , eprinttype =

    Pith/arXiv arXiv

  19. [19]

    RecLM: Recommendation Instruction Tuning

    Jiang, Yangqin and Yang, Yuhao and Xia, Lianghao and Luo, Da and Lin, Kangyi and Huang, Chao , date =. doi:10.48550/arXiv.2412.19302 , url =. 2412.19302 , eprinttype =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2412.19302

  20. [20]

    Survey of

    Ji, Ziwei and Lee, Nayeon and Frieske, Rita and Yu, Tiezheng and Su, Dan and Xu, Yan and Ishii, Etsuko and Bang, Yejin and Chen, Delong and Dai, Wenliang and Chan, Ho Shu and Madotto, Andrea and Fung, Pascale , date =. Survey of. doi:10.1145/3571730 , url =. 2202.03629 , eprinttype =

    Pith/arXiv arXiv

  21. [21]

    Kang, Wang-Cheng and McAuley, Julian , date =. Self-. doi:10.48550/arXiv.1808.09781 , url =. 1808.09781 , eprinttype =

    Pith/arXiv arXiv doi:10.48550/arxiv.1808.09781

  22. [22]

    Conversational Recommender Systems with Large Language Models:

    Kim, Yeong-Hyeon and Cho, Yoon-Sik , date =. Conversational Recommender Systems with Large Language Models:. doi:10.1016/j.eswa.2026.132508 , url =

    arXiv 2026

  23. [23]

    Koren, Yehuda and Bell, Robert and Volinsky, Chris , date =. Matrix. doi:10.1109/MC.2009.263 , url =

    doi:10.1109/mc.2009.263 2009

  24. [24]

    Retrieval-

    Lewis, Patrick and Perez, Ethan and Piktus, Aleksandra and Petroni, Fabio and Karpukhin, Vladimir and Goyal, Naman and Küttler, Heinrich and Lewis, Mike and Yih, Wen-tau and Rocktäschel, Tim and Riedel, Sebastian and Kiela, Douwe , date =. Retrieval-. doi:10.48550/arXiv.2005.11401 , url =. 2005.11401 , eprinttype =

    Pith/arXiv arXiv doi:10.48550/arxiv.2005.11401 2005

  25. [25]

    Liu, Junling and Liu, Chao and Zhou, Peilin and Lv, Renjie and Zhou, Kang and Zhang, Yan , date =. Is. doi:10.48550/arXiv.2304.10149 , url =. 2304.10149 , eprinttype =

    Pith/arXiv arXiv doi:10.48550/arxiv.2304.10149

  26. [26]

    and Lin, Kevin and Hewitt, John and Paranjape, Ashwin and Bevilacqua, Michele and Petroni, Fabio and Liang, Percy , date =

    Liu, Nelson F. and Lin, Kevin and Hewitt, John and Paranjape, Ashwin and Bevilacqua, Michele and Petroni, Fabio and Liang, Percy , date =. Lost in the. doi:10.48550/arXiv.2307.03172 , url =. 2307.03172 , eprinttype =

    Pith/arXiv arXiv doi:10.48550/arxiv.2307.03172

  27. [27]

    Content-Based

    Lops, Pasquale and de Gemmis, Marco and Semeraro, Giovanni , editor =. Content-Based. Recommender. doi:10.1007/978-0-387-85820-3_3 , url =

    doi:10.1007/978-0-387-85820-3_3

  28. [28]

    doi:10.48550/arXiv.2312.16018 , url =

    Luo, Sichun and He, Bowei and Zhao, Haohan and Shao, Wei and Qi, Yanlin and Huang, Yinya and Zhou, Aojun and Yao, Yuxuan and Li, Zongpeng and Xiao, Yuanzhang and Zhan, Mingjie and Song, Linqi , date =. doi:10.48550/arXiv.2312.16018 , url =. 2312.16018 , eprinttype =

    Pith/arXiv arXiv doi:10.48550/arxiv.2312.16018

  29. [29]

    Unifying

    Pan, Shirui and Luo, Linhao and Wang, Yufei and Chen, Chen and Wang, Jiapu and Wu, Xindong , date =. Unifying. doi:10.1109/TKDE.2024.3352100 , url =. 2306.08302 , eprinttype =

    arXiv 2024

  30. [30]

    Perozzi, Bryan and Fatemi, Bahare and Zelle, Dustin and Tsitsulin, Anton and Kazemi, Mehran and Al-Rfou, Rami and Halcrow, Jonathan , date =. Let. doi:10.48550/arXiv.2402.05862 , url =. 2402.05862 , eprinttype =

    Pith/arXiv arXiv doi:10.48550/arxiv.2402.05862

  31. [31]

    Proceedings of the 1994

    Resnick, Paul and Iacovou, Neophytos and Suchak, Mitesh and Bergstrom, Peter and Riedl, John , date =. Proceedings of the 1994. doi:10.1145/192844.192905 , url =

    arXiv 1994

  32. [32]

    doi:10.48550/arXiv.1904.06690 , url =

    Sun, Fei and Liu, Jun and Wu, Jian and Pei, Changhua and Lin, Xiao and Ou, Wenwu and Jiang, Peng , date =. doi:10.48550/arXiv.1904.06690 , url =. 1904.06690 , eprinttype =

    Pith/arXiv arXiv doi:10.48550/arxiv.1904.06690 1904

  33. [33]

    Proceedings of the 25th

    Wang, Xiang and He, Xiangnan and Cao, Yixin and Liu, Meng and Chua, Tat-Seng , date =. Proceedings of the 25th. doi:10.1145/3292500.3330989 , url =. 1905.07854 , eprinttype =

    arXiv 1905

  34. [34]

    Knowledge

    Wang, Shijie and Fan, Wenqi and Feng, Yue and Lin, Shanru and Ma, Xinyu and Wang, Shuaiqiang and Yin, Dawei , date =. Knowledge. doi:10.48550/arXiv.2501.02226 , url =. 2501.02226 , eprinttype =

    Pith/arXiv arXiv doi:10.48550/arxiv.2501.02226

  35. [35]

    Wang, Xiang and He, Xiangnan and Wang, Meng and Feng, Fuli and Chua, Tat-Seng , date =. Neural. Proceedings of the 42nd. doi:10.1145/3331184.3331267 , url =. 1905.08108 , eprinttype =

    arXiv 1905

  36. [36]

    Proceedings of the 27th

    Wang, Hongwei and Zhang, Fuzheng and Wang, Jialin and Zhao, Miao and Li, Wenjie and Xie, Xing and Guo, Minyi , date =. Proceedings of the 27th. doi:10.1145/3269206.3271739 , url =. 1803.03467 , eprinttype =

    arXiv

  37. [37]

    Wei, Suyun and Ye, Ning and Zhang, Shuo and Huang, Xia and Zhu, Jian , date =. Item-. 2012. doi:10.1109/CSSS.2012.507 , url =

    doi:10.1109/csss.2012.507 2012

  38. [38]

    LLMRec: Large Language Models with Graph Augmentation for Recommendation

    Wei, Wei and Ren, Xubin and Tang, Jiabin and Wang, Qinyong and Su, Lixin and Cheng, Suqi and Wang, Junfeng and Yin, Dawei and Huang, Chao , date =. doi:10.48550/arXiv.2311.00423 , url =. 2311.00423 , eprinttype =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2311.00423

  39. [39]

    doi:10.48550/arXiv.2309.05519 , url =

    Wu, Shengqiong and Fei, Hao and Qu, Leigang and Ji, Wei and Chua, Tat-Seng , date =. doi:10.48550/arXiv.2309.05519 , url =. 2309.05519 , eprinttype =

    Pith/arXiv arXiv doi:10.48550/arxiv.2309.05519

  40. [40]

    Retrieve-

    Wu, Yike and Hu, Nan and Bi, Sheng and Qi, Guilin and Ren, Jie and Xie, Anhuan and Song, Wei , date =. Retrieve-

  41. [41]

    Self-Supervised

    Wu, Jiancan and Wang, Xiang and Feng, Fuli and He, Xiangnan and Chen, Liang and Lian, Jianxun and Xie, Xing , date =. Self-Supervised. Proceedings of the 44th. doi:10.1145/3404835.3462862 , url =. 2010.10783 , eprinttype =

    arXiv 2010

  42. [42]

    and Leskovec, Jure , date =

    Ying, Rex and He, Ruining and Chen, Kaifeng and Eksombatchai, Pong and Hamilton, William L. and Leskovec, Jure , date =. Graph. Proceedings of the 24th. doi:10.1145/3219819.3219890 , url =. 1806.01973 , eprinttype =

    arXiv

  43. [43]

    Recommendation as

    Zhang, Junjie and Xie, Ruobing and Hou, Yupeng and Zhao, Wayne Xin and Lin, Leyu and Wen, Ji-Rong , date =. Recommendation as. doi:10.48550/arXiv.2305.07001 , url =. 2305.07001 , eprinttype =

    Pith/arXiv arXiv doi:10.48550/arxiv.2305.07001

  44. [44]

    Retrieval-

    Zhao, Penghao and Zhang, Hailin and Yu, Qinhan and Wang, Zhengren and Geng, Yunteng and Fu, Fangcheng and Yang, Ling and Zhang, Wentao and Jiang, Jie and Cui, Bin , date =. Retrieval-. doi:10.48550/arXiv.2402.19473 , url =. 2402.19473 , eprinttype =

    Pith/arXiv arXiv doi:10.48550/arxiv.2402.19473

  45. [45]

    , title =

    Fortunato, S. , title =. Phys. Rep.-Rev. Sec. Phys. Lett. , volume =. 2010 , pages =

    2010

  46. [46]

    Newman, M. E. J. and Girvan, M. , title =. Phys. Rev. E. , volume =. 2004 , pages =

    2004

  47. [47]

    and Reinhardt, T

    Vehlow, C. and Reinhardt, T. and Weiskopf, D. , title =. IEEE Trans. Vis. Comput. Graph. , volume =. 2013 , pages =

    2013

  48. [48]

    and Albert, R

    Raghavan, U. and Albert, R. and Kumara, S. , title =. Phys. Rev E. , volume =. 2007 , pages =

    2007

  49. [49]

    2011 , pages =

    Robust network community detection using balanced propagation , journal =. 2011 , pages =

    2011

  50. [50]

    and Li, S

    Lou, H. and Li, S. and Zhao, Y. , title =. Physica A. , volume =. 2013 , pages =

    2013

  51. [51]

    and Newman, M

    Clauset, A. and Newman, M. E. J. and Moore, C. , title =. Phys. Rev. E. , volume =. 2004 , pages =

    2004

  52. [52]

    Blondel, V. D. and Guillaume, J. L. and Lambiotte, R. and Lefebvre, E. , title =. J. Stat. Mech.-Theory Exp. , volume =. 2008 , pages =

    2008

  53. [53]

    and Campari, R

    Sobolevsky, S. and Campari, R. , title =. Phys. Rev. E. , volume =. 2014 , pages =

    2014

  54. [54]

    and Barthelemy, M

    Fortunato, S. and Barthelemy, M. , title =. Proc. Natl. Acad. Sci. U. S. A. , volume =. 2007 , pages =

    2007

  55. [55]

    2011 , pages =

    Unfolding communities in large complex networks: Combining defensive and offensive label propagation for core extraction , journal =. 2011 , pages =

    2011

  56. [56]

    and Li, J

    Wang, X. and Li, J. , title =. Physica A. , volume =. 2013 , pages =

    2013

  57. [57]

    and Wang, X

    Li, J. and Wang, X. and Eustace, J. , title =. Physica A. , volume =. 2013 , pages =

    2013

  58. [58]

    Fabio, D. R. and Fabio, D. and Carlo, P. , title =. Sci. Rep. , volume =. 2013 , pages =

    2013

  59. [59]

    and Wu, T

    Chen, Q. and Wu, T. T. and Fang, M. , title =. Physica A. , volume =. 2013 , pages =

    2013

  60. [60]

    and Wang, R

    Zhang, S. and Wang, R. and Zhang, X. , title =. Physica A. , volume =. 2007 , pages =

    2007

  61. [61]

    and Petr\'oczi, A

    Nepusz, T. and Petr\'oczi, A. and N\'egyessy, L. and Bazs\'o, F. , title =. Phys. Rev. E. , volume =. 2008 , pages =

    2008

  62. [62]

    and Liang, Z

    Fabricio, B. and Liang, Z. , title =. Soft Comput. , volume =. 2013 , pages =

    2013

  63. [63]

    and Gao, L

    Sun, P. and Gao, L. and Han, S. , title =. Inf. Sci. , volume =. 2011 , pages =

    2011

  64. [64]

    and Liu, D

    Wang, W. and Liu, D. and Liu, X. and Pan, L. , title =. Physica A. , volume =. 2013 , pages =

    2013

  65. [65]

    and Roberts, S

    Psorakis, I. and Roberts, S. and Ebden, M. and Sheldon, B. , title =. Phys. Rev. E. , volume =. 2011 , pages =

    2011

  66. [66]

    and Yeung, D

    Zhang, Y. and Yeung, D. , title =. In Proc. ACM SIGKDD Conf. , year =

  67. [67]

    , title =

    Liu, J. , title =. Eur. Phys. J. B. , volume =. 2010 , pages =

    2010

  68. [68]

    Havens, T. C. and Bezdek, J. C. and Leckie, C. and Ramamohanarao, K. and Palaniswami, M. , title =. IEEE Trans. Fuzzy Syst. , volume =. 2013 , pages =

    2013

  69. [69]

    Newman, M. E. J. , title =

  70. [70]

    2012 , pages =

    Ubiquitousness of link-density and link-pattern communities in real-world networks , journal =. 2012 , pages =

    2012

  71. [71]

    and Fortunato, S

    Lancichinetti, A. and Fortunato, S. and Radicchi, F. , title =. Phys. Rev. E. , volume =. 2008 , pages =

    2008

  72. [72]

    and Pellegrini, M

    Liu, W. and Pellegrini, M. and Wang, X. , title =. Sci. Rep. , volume =. 2014 , pages =

    2014

  73. [73]

    and Diaz-Guilera, A

    Danon, L. and Diaz-Guilera, A. and Duch, J. and Arenas, A. , title =. J. Stat. Mech.-Theory Exp. , volume =. 2005 , pages =

    2005

  74. [74]

    , title =

    Gregory, S. , title =. J. Stat. Mech.-Theory Exp. , volume =. 2011 , pages =

    2011

  75. [75]

    and Fortunato, S

    Lancichinetti, A. and Fortunato, S. , title =. Phys. Rev. E. , volume =. 2009 , pages =

    2009

  76. [76]

    and Rifqi, M

    Hullermeier, E. and Rifqi, M. , title =. in Proc. IFSA/EUSFLAT Conf. , year =