pith. sign in

arxiv: 2602.18249 · v2 · pith:2IYNNZXYnew · submitted 2026-02-20 · 💻 cs.IR

Dual-Tree LLM-Enhanced Negative Sampling for Implicit Collaborative Filtering

Jiayi Wu , Zhengyu Wu , Xunkai Li , Rong-Hua Li , Guoren Wang This is my paper

Pith reviewed 2026-05-21 12:32 UTC · model grok-4.3

classification 💻 cs.IR
keywords negative samplingimplicit collaborative filteringlarge language modelsrecommendation systemsfalse negative identificationhierarchical index treeshard negative miningtext-free LLM inference
0
0 comments X

The pith

A dual-tree method turns collaborative data into item-ID encodings that let LLMs identify false negatives for training without text or fine-tuning.

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

The paper sets out to show that negative sampling in implicit collaborative filtering can be strengthened by feeding large language models structured encodings built from user-item interaction graphs rather than item descriptions. Hierarchical index trees first organize both structural connections and latent patterns into compact item-ID sequences. An off-the-shelf LLM then uses these sequences to flag false negatives among unobserved items, after which a multi-view sampler blends user preference scores with item-item similarities drawn from the same encodings. If the claim holds, training signals become cleaner for a wide range of existing recommenders, raising accuracy while removing the usual requirements for textual metadata and task-specific model updates.

Core claim

The authors claim that hierarchical index trees can encode collaborative structural information and latent semantic patterns into ordered item-ID sequences that enable an unmodified large language model to distinguish false negatives from true negatives in implicit feedback data; combining the resulting identifications with multi-view item similarities then yields harder and more informative negative samples that improve the discriminative power of implicit collaborative filtering models.

What carries the argument

The dual-tree LLM-enhanced negative sampling (DTL-NS) pipeline, which first builds hierarchical index trees to produce text-free structured item-ID encodings for LLM-based false-negative detection and then fuses those detections with user-item scores and item-item similarities for multi-view hard-negative mining.

If this is right

  • Implicit CF models trained with DTL-NS produce higher ranking accuracy on held-out interactions than the same models trained with conventional negative sampling.
  • The same LLM-based false-negative module works without changes when plugged into multiple existing CF architectures and alternative negative sampling heuristics.
  • Performance gains remain stable when different LLMs are substituted, indicating that the structured encodings rather than any single model drive the improvement.
  • The offline tree construction step adds negligible online cost while supplying reusable encodings that support repeated sampling rounds during training.

Where Pith is reading between the lines

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

  • The same encoding strategy could be tested on graph-based tasks outside recommendation, such as link prediction, where only structural data is available.
  • If the LLM judgments prove stable across datasets, the approach might lower the barrier to using large models in domains that lack rich textual metadata.
  • Extending the trees to incorporate temporal interaction order could further refine the notion of what counts as a false negative in evolving user histories.

Load-bearing premise

The method rests on the premise that hierarchical index trees can convert user-item interaction patterns into item-ID sequences that allow an LLM to reliably detect false negatives even when no textual item information or extra training is supplied.

What would settle it

A side-by-side run on the same dataset in which one model uses standard negative sampling and another uses the dual-tree encodings fed to the LLM; if the latter shows no consistent lift in ranking metrics such as NDCG or Recall@K, the central claim is refuted.

Figures

Figures reproduced from arXiv: 2602.18249 by Guoren Wang, Jiayi Wu, Rong-Hua Li, Xunkai Li, Zhengyu Wu.

Figure 1
Figure 1. Figure 1: LLM-based false-negative identification accuracy. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: The DTL-NS framework. We depict a single tree-based item encoding example, since the dual trees use node embeddings [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 4
Figure 4. Figure 4: The impact of 𝛼𝑐 and 𝛼𝑠 . Amazon-sports, Amazon-toys, and Yelp. The average per-epoch forward-pass time increases slightly from 1.26s, 1.51s, and 1.93s to 1.57s, 1.73s, and 2.47s, indicating that the additional overhead of multi-view negative selection is marginal. Finally, we compare the overall training efficiency of DTL-NS with negative sampling baselines. As shown in [PITH_FULL_IMAGE:figures/full_fig_… view at source ↗
Figure 5
Figure 5. Figure 5: The impact of different variant methods. [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Performance comparison of integrating DTL-NS [PITH_FULL_IMAGE:figures/full_fig_p009_6.png] view at source ↗
read the original abstract

Negative sampling is a pivotal technique in implicit collaborative filtering (CF) recommendation, enabling efficient and effective training by contrasting observed interactions with sampled unobserved ones. Recently, large language models (LLMs) have shown promise in recommender systems; however, research on LLM-empowered negative sampling remains underexplored. Existing methods heavily rely on textual information and task-specific fine-tuning, limiting practical applicability. To this end, we propose a text-free and fine-tuning-free Dual-Tree LLM-enhanced Negative Sampling method (DTL-NS). It consists of two modules: (i) an offline false negative identification module that leverages hierarchical index trees to transform collaborative structural and latent semantic information into structured item-ID encodings for LLM inference, enabling accurate identification of false negatives; and (ii) a multi-view hard negative sampling module that combines user-item preference scores with item-item hierarchical similarities from these encodings to mine high-quality negatives, thus improving the discriminative ability of recommender models. Extensive experiments demonstrate the effectiveness of DTL-NS. Moreover, DTL-NS shows broad applicability across different implicit CF models, negative sampling methods, and LLMs, consistently enhancing recommendation performance.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 2 minor

Summary. The manuscript proposes Dual-Tree LLM-Enhanced Negative Sampling (DTL-NS) for implicit collaborative filtering. It comprises an offline false-negative identification module that encodes collaborative structure and latent semantics into hierarchical item-ID strings for an off-the-shelf LLM to detect false negatives without textual metadata or task-specific fine-tuning, plus a multi-view hard-negative sampler that fuses user-item preference scores with item-item similarities derived from the same encodings. The authors assert that extensive experiments confirm effectiveness and broad applicability across CF models, negative samplers, and LLMs.

Significance. If the central premise holds—that LLMs can extract non-random, semantically useful signals from purely structural ID encodings—this would offer a practical advance by removing dependence on item text and fine-tuning, potentially simplifying deployment of LLM-assisted negative sampling in production recommenders. Demonstrated cross-model and cross-LLM gains would further increase its utility for the field.

major comments (3)
  1. [§3.1] §3.1 (False Negative Identification Module): The claim that hierarchical index trees produce encodings enabling an LLM to accurately distinguish false negatives from true negatives with zero textual information and zero fine-tuning is load-bearing for both modules. No ablation or control experiment is described that isolates whether LLM decisions arise from semantic content versus tokenization artifacts, positional bias, or ID string length.
  2. [§4] §4 (Experiments): The broad-applicability claim requires evidence that performance gains survive standard controls (different random seeds, temporal vs. random splits, and multiple base negative samplers). Without reported statistical significance or variance across runs, it is unclear whether the observed improvements are robust or attributable to the LLM component versus the tree-based similarity alone.
  3. [§3.2] §3.2 (Multi-view Hard Negative Sampling): The integration of LLM-derived false-negative labels with item-item hierarchical similarities is presented as additive, yet no analysis shows that the LLM labels contribute incremental signal beyond what the tree similarities already provide; an ablation removing the LLM step would directly test this.
minor comments (2)
  1. [Figure 1] Figure 1 (system overview) would benefit from explicit annotation of the two tree structures and the exact format of the item-ID strings passed to the LLM.
  2. [§3] Notation for the dual-tree construction (e.g., definitions of parent-child relations and encoding functions) should be introduced with a small worked example to improve readability.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for their insightful comments on our manuscript. We have carefully considered each point and provide detailed responses below. Where appropriate, we will revise the manuscript to address the concerns raised.

read point-by-point responses

  1. Referee: [§3.1] §3.1 (False Negative Identification Module): The claim that hierarchical index trees produce encodings enabling an LLM to accurately distinguish false negatives from true negatives with zero textual information and zero fine-tuning is load-bearing for both modules. No ablation or control experiment is described that isolates whether LLM decisions arise from semantic content versus tokenization artifacts, positional bias, or ID string length.

    Authors: We agree that demonstrating the source of the LLM's discriminative ability is crucial. To isolate the effect of the hierarchical index trees, we will add a new ablation experiment in the revised manuscript. Specifically, we will compare the false negative identification performance using the structured hierarchical ID encodings against control conditions such as randomly generated ID strings of similar length and format, as well as shuffled versions of the hierarchical encodings. This will help confirm that the LLM is leveraging the encoded collaborative structure and latent semantics rather than superficial artifacts like tokenization or string length. revision: yes

  2. Referee: [§4] §4 (Experiments): The broad-applicability claim requires evidence that performance gains survive standard controls (different random seeds, temporal vs. random splits, and multiple base negative samplers). Without reported statistical significance or variance across runs, it is unclear whether the observed improvements are robust or attributable to the LLM component versus the tree-based similarity alone.

    Authors: We acknowledge the importance of these robustness checks for establishing broad applicability. In the revised version, we will expand the experimental section to include: (1) results averaged over multiple random seeds with reported standard deviations, (2) statistical significance tests (such as paired t-tests with p-values) comparing DTL-NS against baselines, (3) evaluations on both random and temporal data splits, and (4) additional experiments using at least two other base negative sampling methods to demonstrate compatibility. These additions will clarify the contribution of the LLM component and the overall robustness of the gains. revision: yes

  3. Referee: [§3.2] §3.2 (Multi-view Hard Negative Sampling): The integration of LLM-derived false-negative labels with item-item hierarchical similarities is presented as additive, yet no analysis shows that the LLM labels contribute incremental signal beyond what the tree similarities already provide; an ablation removing the LLM step would directly test this.

    Authors: We concur that an ablation isolating the LLM-derived labels would provide valuable insight into their incremental value. We will incorporate an ablation study in Section 3.2 and the experiments, where we evaluate a variant of the multi-view sampler that uses only the item-item hierarchical similarities without incorporating the false-negative labels from the LLM. By comparing this to the full model, we will quantify the additional signal provided by the LLM component. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation is self-contained with independent modules and external experimental validation

full rationale

The paper introduces DTL-NS as two distinct engineering modules: an offline false-negative identification step that encodes collaborative structure into item-ID strings via hierarchical trees, and a multi-view hard-negative sampler that combines preference scores with hierarchical similarities. Effectiveness is asserted via experiments on multiple implicit CF models, sampling methods, and LLMs rather than any closed-form derivation or fitted parameter that is then relabeled as a prediction. No equations, uniqueness theorems, or ansatzes are shown to reduce to prior self-citations or internal fits by construction. The central claim therefore rests on empirical outcomes outside the method definition itself, qualifying as a normal non-circular engineering contribution.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review supplies insufficient technical detail to enumerate concrete free parameters, axioms, or invented entities; the central claim rests on the unstated premise that LLM inference on tree-derived ID strings can substitute for textual semantics.

pith-pipeline@v0.9.0 · 5741 in / 1112 out tokens · 39938 ms · 2026-05-21T12:32:07.061729+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

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

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

Reference graph

Works this paper leans on

52 extracted references · 52 canonical work pages · 1 internal anchor

  1. [1]

    Mohiuddin Ahmed, Raihan Seraj, and Syed Mohammed Shamsul Islam. 2020. The k-means algorithm: A comprehensive survey and performance evaluation. Electronics9, 8 (2020), 1295

    work page 2020

  2. [2]

    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. InProceedings of the 17th ACM conference on recommender systems. 1007–1014

    work page 2023

  3. [3]

    Yao Cai, Weimao Ke, Eric Cui, and Fei Yu. 2022. A deep recommendation model of cross-grained sentiments of user reviews and ratings.Information Processing & Management59, 2 (2022), 102842

    work page 2022

  4. [4]

    Chong Chen, Weizhi Ma, Min Zhang, Chenyang Wang, Yiqun Liu, and Shaoping Ma. 2023. Revisiting negative sampling vs. non-sampling in implicit recommen- dation.ACM Transactions on Information Systems41, 1 (2023), 1–25

    work page 2023

  5. [5]

    Lei Chen, Chen Gao, Xiaoyi Du, Hengliang Luo, Depeng Jin, Yong Li, and Meng Wang. 2025. Enhancing ID-based recommendation with large language models. ACM Transactions on Information Systems43, 5 (2025), 1–30

    work page 2025

  6. [6]

    Jingtao Ding, Yuhan Quan, Quanming Yao, Yong Li, and Depeng Jin. 2020. Simplify and robustify negative sampling for implicit collaborative filtering.Advances in Neural Information Processing Systems33 (2020), 1094–1105

    work page 2020

  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). InProceedings of the 16th ACM conference on recommender systems. 299–315

    work page 2022

  8. [8]

    Xavier Glorot and Yoshua Bengio. 2010. Understanding the difficulty of training deep feedforward neural networks. InProceedings of the thirteenth international conference on artificial intelligence and statistics. JMLR Workshop and Conference Proceedings, 249–256

    work page 2010

  9. [9]

    Aaron Grattafiori, Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Alex Vaughan, et al. 2024. The llama 3 herd of models.arXiv preprint arXiv:2407.21783 (2024)

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  10. [10]

    Xiangnan He, Kuan Deng, Xiang Wang, Yan Li, Yongdong Zhang, and Meng Wang. 2020. Lightgcn: Simplifying and powering graph convolution network for recommendation. InProceedings of the 43rd International ACM SIGIR conference on research and development in Information Retrieval. 639–648

    work page 2020

  11. [11]

    Xiangnan He, Lizi Liao, Hanwang Zhang, Liqiang Nie, Xia Hu, and Tat-Seng Chua. 2017. Neural collaborative filtering. InProceedings of the 26th international conference on world wide web. 173–182

    work page 2017

  12. [12]

    Henderi Henderi, Tri Wahyuningsih, and Efana Rahwanto. 2021. Comparison of Min-Max normalization and Z-Score Normalization in the K-nearest neighbor (kNN) Algorithm to Test the Accuracy of Types of Breast Cancer.International Journal of Informatics and Information Systems4, 1 (2021), 13–20

    work page 2021

  13. [13]

    Wenyue Hua, Shuyuan Xu, Yingqiang Ge, and Yongfeng Zhang. 2023. How to index item ids for recommendation foundation models. InProceedings of the Annual International ACM SIGIR Conference on Research and Development in Information Retrieval in the Asia Pacific Region. 195–204

    work page 2023

  14. [14]

    Tinglin Huang, Yuxiao Dong, Ming Ding, Zhen Yang, Wenzheng Feng, Xinyu Wang, and Jie Tang. 2021. Mixgcf: An improved training method for graph neural network-based recommender systems. InProceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining. 665–674

    work page 2021

  15. [15]

    Christopher C Johnson et al . 2014. Logistic matrix factorization for implicit feedback data.Advances in Neural Information Processing Systems27, 78 (2014), 1–9

    work page 2014

  16. [16]

    Juha Kärkkäinen, Marcin Piątkowski, and Simon J Puglisi. 2017. String inference from longest-common-prefix array. InInternational Colloquium on Automata, Languages and Programming. Schloss Dagstuhl-Leibniz-Zentrum für Informatik, 62

    work page 2017

  17. [17]

    Diederik P Kingma and Jimmy Ba. 2015. Adam: A method for stochastic opti- mization. InThe 3rd International Conference on Learning Representations

    work page 2015

  18. [18]

    Woosuk Kwon, Zhuohan Li, Siyuan Zhuang, Ying Sheng, Lianmin Zheng, Cody Hao Yu, Joseph Gonzalez, Hao Zhang, and Ion Stoica. 2023. Efficient memory management for large language model serving with pagedattention. In Proceedings of the 29th symposium on operating systems principles. 611–626

    work page 2023

  19. [19]

    Riwei Lai, Li Chen, Yuhan Zhao, Rui Chen, and Qilong Han. 2023. Disentangled negative sampling for collaborative filtering. InProceedings of the Sixteenth ACM International Conference on Web Search and Data Mining. 96–104

    work page 2023

  20. [20]

    Riwei Lai, Rui Chen, Qilong Han, Chi Zhang, and Li Chen. 2024. Adaptive hardness negative sampling for collaborative filtering. InProceedings of the AAAI Conference on Artificial Intelligence, Vol. 38. 8645–8652

    work page 2024

  21. [21]

    Xinhang Li, Chong Chen, Xiangyu Zhao, Yong Zhang, and Chunxiao Xing. 2024. E4SRec: An Elegant Effective Efficient Extensible Solution of Large Language Models for Sequential Recommendation. InProceedings of The Web Conference

    work page 2024

  22. [22]

    George Ma, Yifei Wang, and Yisen Wang. 2023. Laplacian canonization: A min- imalist approach to sign and basis invariant spectral embedding.Advances in Neural Information Processing Systems36 (2023), 11296–11337

    work page 2023

  23. [23]

    Haokai Ma, Ruobing Xie, Lei Meng, Xin Chen, Xu Zhang, Leyu Lin, and Jie Zhou

    work page

  24. [24]

    In Proceedings of the 17th ACM conference on recommender systems

    Exploring false hard negative sample in cross-domain recommendation. In Proceedings of the 17th ACM conference on recommender systems. 502–514

    work page

  25. [25]

    Suphakit Niwattanakul, Jatsada Singthongchai, Ekkachai Naenudorn, and Su- pachanun Wanapu. 2013. Using of Jaccard coefficient for keywords similarity. InProceedings of the international multiconference of engineers and computer scientists, Vol. 1. 380–384

    work page 2013

  26. [26]

    Xubin Ren, Wei Wei, Lianghao Xia, Lixin Su, Suqi Cheng, Junfeng Wang, Dawei Yin, and Chao Huang. 2024. Representation learning with large language models for recommendation. InProceedings of the ACM web conference 2024. 3464–3475

    work page 2024

  27. [27]

    Steffen Rendle, Christoph Freudenthaler, Zeno Gantner, and Lars Schmidt-Thieme

    work page

  28. [28]

    InProceedings of the 25th Conference on Uncertainty in Artificial Intelligence

    BPR: Bayesian personalized ranking from implicit feedback. InProceedings of the 25th Conference on Uncertainty in Artificial Intelligence

    work page

  29. [29]

    Leheng Sheng, An Zhang, Yi Zhang, Yuxin Chen, Xiang Wang, and Tat-Seng Chua. [n. d.]. Language Representations Can be What Recommenders Need: Findings and Potentials. InThe Thirteenth International Conference on Learning Representations

    work page

  30. [30]

    Wentao Shi, Jiawei Chen, Fuli Feng, Jizhi Zhang, Junkang Wu, Chongming Gao, and Xiangnan He. 2023. On the theories behind hard negative sampling for recommendation. InProceedings of the ACM Web Conference 2023. 812–822

    work page 2023

  31. [31]

    Xiang Wang, Xiangnan He, Meng Wang, Fuli Feng, and Tat-Seng Chua. 2019. Neural graph collaborative filtering. InProceedings of the 42nd international ACM SIGIR conference on Research and development in Information Retrieval. 165–174

    work page 2019

  32. [32]

    Wei Wei, Xubin Ren, Jiabin Tang, Qinyong Wang, Lixin Su, Suqi Cheng, Jun- feng Wang, Dawei Yin, and Chao Huang. 2024. Llmrec: Large language models with graph augmentation for recommendation. InProceedings of the 17th ACM international conference on web search and data mining. 806–815

    work page 2024

  33. [33]

    Chuhan Wu, Fangzhao Wu, Yongfeng Huang, and Xing Xie. 2023. Personal- ized news recommendation: Methods and challenges.ACM Transactions on Information Systems41, 1 (2023), 1–50

    work page 2023

  34. [34]

    Yunjia Xi, Weiwen Liu, Jianghao Lin, Xiaoling Cai, Hong Zhu, Jieming Zhu, Bo Chen, Ruiming Tang, Weinan Zhang, and Yong Yu. 2024. Towards open-world recommendation with knowledge augmentation from large language models. In Proceedings of the 18th ACM Conference on Recommender Systems. 12–22

    work page 2024

  35. [35]

    Xin Xin, Xiangyuan Liu, Hanbing Wang, Pengjie Ren, Zhumin Chen, Jiahuan Lei, Xinlei Shi, Hengliang Luo, Joemon M Jose, Maarten de Rijke, et al . 2023. Improving implicit feedback-based recommendation through multi-behavior alignment. InProceedings of the 46th international ACM SIGIR conference on research and development in information retrieval. 932–941

    work page 2023

  36. [36]

    Wujiang Xu, Zujie Liang, Jiaojiao Han, Xuying Ning, Wenfang Lin, Linxun Chen, Feng Wei, and Yongfeng Zhang. 2025. SLMRec: Distilling Large Language Models into Small for Sequential Recommendation. InInternational Conference on Learning Representations (ICLR 2025)

    work page 2025

  37. [37]

    Zhen Yang, Ming Ding, Tinglin Huang, Yukuo Cen, Junshuai Song, Bin Xu, Yuxiao Dong, and Jie Tang. 2024. Does negative sampling matter? a review with insights into its theory and applications.IEEE Transactions on Pattern Analysis Dual-Tree LLM-Enhanced Negative Sampling for Implicit Collaborative Filtering Conference’17, July 2017, Washington, DC, USA and ...

    work page 2024

  38. [38]

    Zhen Yang, Ming Ding, Chang Zhou, Hongxia Yang, Jingren Zhou, and Jie Tang

    work page

  39. [39]

    InPro- ceedings of the 26th ACM SIGKDD international conference on knowledge discovery & data mining

    Understanding negative sampling in graph representation learning. InPro- ceedings of the 26th ACM SIGKDD international conference on knowledge discovery & data mining. 1666–1676

    work page

  40. [40]

    Zhen Yang, Ming Ding, Xu Zou, Jie Tang, Bin Xu, Chang Zhou, and Hongxia Yang. 2022. Region or global? A principle for negative sampling in graph-based recommendation.IEEE Transactions on Knowledge and Data Engineering35, 6 (2022), 6264–6277

    work page 2022

  41. [41]

    Jianwen Yin, Chenghao Liu, Jundong Li, BingTian Dai, Yun-chen Chen, Min Wu, and Jianling Sun. 2019. Online collaborative filtering with implicit feedback. In International Conference on Database Systems for Advanced Applications. Springer, 433–448

    work page 2019

  42. [42]

    Chunxu Zhang, Guodong Long, Zijian Zhang, Zhiwei Li, Honglei Zhang, Qiang Yang, and Bo Yang. 2025. Personalized recommendation models in federated settings: A survey.IEEE Transactions on Knowledge and Data Engineering(2025)

    work page 2025

  43. [43]

    Hongyi Zhang, Moustapha Cisse, Yann N Dauphin, and David Lopez-Paz. 2018. mixup: Beyond empirical risk minimization. InProceedings of the 6th International Conference on Learning Representations

    work page 2018

  44. [44]

    Junjie Zhang, Ruobing Xie, Yupeng Hou, Xin Zhao, Leyu Lin, and Ji-Rong Wen

    work page

  45. [45]

    Recommendation as instruction following: A large language model em- powered recommendation approach.ACM Transactions on Information Systems 43, 5 (2025), 1–37

    work page 2025

  46. [46]

    Weinan Zhang, Tianqi Chen, Jun Wang, and Yong Yu. 2013. Optimizing top-n collaborative filtering via dynamic negative item sampling. InProceedings of the 36th international ACM SIGIR conference on Research and development in information retrieval. 785–788

    work page 2013

  47. [47]

    Yang Zhang, Keqin Bao, Ming Yan, Wenjie Wang, Fuli Feng, and Xiangnan He

    work page

  48. [48]

    InProceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)

    Text-like Encoding of Collaborative Information in Large Language Models for Recommendation. InProceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 9181–9191

    work page

  49. [49]

    Chu Zhao, Enneng Yang, Yuting Liu, Jianzhe Zhao, Guibing Guo, and Xingwei Wang. 2025. Can LLM-Driven Hard Negative Sampling Empower Collaborative Filtering? Findings and Potentials.arXiv preprint arXiv:2504.04726(2025)

    work page arXiv 2025

  50. [50]

    Kai Zhao, Yukun Zheng, Tao Zhuang, Xiang Li, and Xiaoyi Zeng. 2022. Joint learning of e-commerce search and recommendation with a unified graph neural network. InProceedings of the Fifteenth ACM International Conference on Web Search and Data Mining. 1461–1469

    work page 2022

  51. [51]

    Yuhan Zhao, Rui Chen, Riwei Lai, Qilong Han, Hongtao Song, and Li Chen. 2023. Augmented Negative Sampling for Collaborative Filtering. InProceedings of the 17th ACM Conference on Recommender Systems. 256–266

    work page 2023

  52. [52]

    Qiannan Zhu, Haobo Zhang, Qing He, and Zhicheng Dou. 2022. A gain-tuning dynamic negative sampler for recommendation. InProceedings of the ACM web conference 2022. 277–285

    work page 2022