Pith. sign in

REVIEW 4 major objections 6 minor 1 cited by

PageLLM: A Multi-Grained Reward Framework for Whole-Page Optimization with Large Language Models

T0 review · 4 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read PageLLM claims implicit user feedback alone suffices for LLM whole-page optimization if the reward splits into page-level and item-level heads; the joint reward raises NDCG@100 by up to 46.8% and lifts GMV by 0.44% online.

desk verdict A clean reward-decoupling idea and a useful preference-pair taxonomy, but the headline claims rest on a circular heuristic and a mislabeled 'implicit feedback' story. read the letter →

arxiv 2506.09084 v2 pith:ZB52IGIN submitted 2025-06-10 cs.LG cs.AI

Xinyuan Wang , Liang Wu , Dongjie Wang , Yanjie Fu This is my paper
classification cs.LGcs.AI
keywords whole-pageoptimizationlargelanguagemodelsRLHFmulti-grainedrewardimplicitfeedbackrecommendersystemsPPOe-commerceranking
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper tries to establish that whole-page optimization for search and recommendation pages can be driven by implicit user feedback alone, without human preference annotations, provided the reward signal is decoupled into two granularities. The proposed PageLLM framework builds a golden ranked list and four contrastive preference-pair families from interaction logs, trains a coarse page-level reward head and a fine item-level reward head on the same pairs, and uses their sum to fine-tune a pre-trained LLM with PPO. The paper's central evidence is an ablation in which removing either reward head reduces NDCG@100 by 15-18% while the joint reward improves it by up to 46.8%, plus a production A/B test with over ten million users that lifts GMV by 0.44% and click-through rate by 0.14%. If these results hold, annotation-free RLHF can scale to industrial whole-page optimization, and reward granularity, not label volume, is the binding constraint.

What carries the argument

The central object is the multi-grained reward, a sum $R=R_c+R_f$ of a coarse page-level head and a fine item-level head. The page-level head $R_c$ is a Bradley-Terry model that scores each generated list as one indivisible unit, so small positional changes are washed out and list-level cues like category mix and coverage dominate. The item-level head $R_f$ averages per-item engagement-sensitive position-swap scores, so a single misplaced item produces a gradient even when the page-level score is saturated. Both heads are trained on the same four preference-pair families (relevance, ranking, diversity, redundancy) through different losses, and this sum is the objective that PPO optimizes against a KL penalty to the supervised reference policy.

What would settle it

Train PageLLM with a single scalar Bradley-Terry reward on the same four preference-pair families and compare NDCG@100 on held-out categories; if the single-scalar model matches the two-head model within seed-to-seed variation, the complementarity claim fails. A second arm builds the golden list from human page rankings instead of the rating/popularity/category recipe; if that variant outperforms the heuristic-trained one, the implicit-feedback result depends on the label recipe.

Watch

Extended reading notes

Core claim

The paper's central discovery is that the granularity of the reward, not the source of supervision, is the limiting factor for LLM-based whole-page optimization. PageLLM constructs a golden ranked list and four families of negative lists from implicit feedback, trains a coarse page-level reward $R_c$ and a fine item-level reward $R_f$ on those same lists, and uses $R=R_c+R_f$ as the PPO objective for a pre-trained LLM. On seven product categories the joint reward is reported to improve NDCG@100 by up to 46.8%, while removing either head alone costs 17.8% or 15.2%, which the paper reads as proof that the two granularities are complementary rather than redundant. In a one-week A/B test with over ten million users the treatment raises GMV by 0.44% and CTR by 0.14%. The claim, stated on the paper's own terms, is that no single scalar reward can subsume both page-level coherence and item-level positional sharpness.

Load-bearing premise

The golden list is assumed to faithfully represent whole-page user satisfaction: if the rating-sorted, popularity tie-broken, category-capped ordering mis-ranks items or gets the diversity-versus-redundancy balance wrong, the reward model learns that distortion and the reported gains measure fit to the label recipe rather than genuine page-quality improvement.

Editorial extensions

If this is right

  • The preference-pair construction is mechanical, so the entire RLHF pipeline can be retrained from interaction logs without human annotation.
  • The ablation's non-additive drops imply that future WPO reward designs should use multiple granularities; a single scalar reward will miss either coherence or positional sharpness.
  • Because the page-level head can be served as a CPU ranking feature before the full LLM policy is deployed, the framework supports incremental rollouts on existing search infrastructure.
  • The gains concentrate in categories where preferences are most differentiated, so the item-level head pays off most where positional signal is rich.
  • Under a 50% cold-start simulation, PageLLM degrades less than the sequential and autoencoder baselines it is compared with, suggesting the coarse reward preserves list-level structure when interaction data is sparse.

Reading between the lines

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

  • A direct extension would replace the hand-built golden list with human page rankings; if the heuristic-trained reward then loses its NDCG gain, the 'implicit feedback suffices' claim would narrow to 'this label recipe suffices'.
  • Because the two heads are added with fixed unit weights, the framework invites learned combination weights or additional granular heads, such as image-quality or price-salience heads, to test whether the super-additive gain continues.
  • The item-level reward models only within-list position swaps; extending it to cross-session or cross-page effects could tighten the small online lifts the paper reports.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper proposes PageLLM, a framework for whole-page optimization (WPO) that uses an LLM to generate ranked recommendation pages. It constructs a per-user 'golden' target list from explicit ratings (r_ui > 3), popularity tie-breaking (Eq. 6), and a category cap (λ=0.4), then derives four preference-pair families (relevance, ranking, diversity, redundancy) from that list. Two reward heads are trained on these pairs: a page-level Bradley–Terry head and an item-level head defined as the average of per-item scores, and PPO fine-tunes a GPT-2/Llama backbone using their sum. Offline results on seven Amazon categories report NDCG@100 gains over eleven baselines, and an ablation on AM-Toys reports that removing either reward head reduces NDCG@100 by 15–18%. An online A/B test with over 10 million users reports GMV +0.44% and CTR +0.14%. The paper's central claims are that implicit feedback alone suffices for alignment and that the two reward granularities are complementary.

Significance. If the central claims held, PageLLM would be a significant advance: a label-free RLHF pipeline for WPO, a multi-grained reward with empirical evidence of complementarity, and a production deployment demonstrating online lift. The paper also contributes a clear decomposition of page quality into four preference-pair families and releases code and data. However, the reward model and the evaluation metric are both built from the same hand-constructed golden list, so the headline gains and the complementarity claim may reflect fitting to that recipe rather than to user satisfaction. The online A/B description is ambiguous about whether the full LLM policy was served. These issues are load-bearing but, in my view, addressable with additional experiments and clarifications.

major comments (4)
  1. [§4.2.3, §4.2.4, §5.1.2] The reward model and the evaluation metric encode the same hand-built page-quality heuristic. The golden list σ_gt is constructed from rating order, popularity tie-breaking (Eq. 6), and category cap λ=0.4; all four preference-pair families in Eq. (3) are generated by corrupting σ_gt; and NDCG@100, the headline metric, is computed against held-out relevance labels derived from the same binarized ratings (r_ui > 3). The 46.8% NDCG gain and the 17.8%/15.2% ablation drops may therefore only show that PageLLM learns the authors' label recipe. A concrete test would be to evaluate against an independently measured engagement signal (clicks, purchases, or dwell time), or at least to vary λ and the rating threshold and show that the conclusions are insensitive to the recipe.
  2. [§4.4.2, Eq. (12)] The item-level reward is asserted to capture 'engagement-sensitive position swaps,' but the engagement delta ΔE(u, i_k → i'_k) in Eq. (12) is never computed from logs. The fine head is trained with the same preference pairs and a pairwise Bradley–Terry loss (Eq. 14); no actual engagement measurement enters the training pipeline. Either provide a concrete operationalization of ΔE from user behavior, or revise the claim that the item-level head reflects engagement rather than the authors' swap-based ordering of the golden list.
  3. [§5.3] The online experiment description is internally inconsistent about what was served. The first paragraph states that 'the treatment served the full PageLLM policy,' but the final paragraph says 'the coarse reward R_c was used as an auxiliary ranking feature during the rollout phase.' If only R_c augmented the incumbent pipeline, the GMV/CTR lifts cannot be attributed to the full PageLLM framework, nor do they provide the independent validation needed to break the circularity of the offline evaluation. Please clarify the exact serving configuration and, if the full policy was not served, label the online result accordingly.
  4. [§5.4, Figure 5] The load-bearing claim that the two reward heads are complementary rests on a single-category ablation (AM-Toys) with no error bars or significance tests, even though §5.1.4 states that reported numbers are averaged over five seeds. The differences among page-only, item-only, and full PageLLM in Figure 5 should be reported with standard deviations or confidence intervals, and the ablation should be run on at least a second category to support the general 'super-additive' conclusion.
minor comments (6)
  1. [Abstract, §1, §4.2.2] Throughout, the paper calls the five-star review scores 'implicit user feedback'; these are explicit ratings. Please either rename the signals or clarify how the framework would use truly implicit signals such as clicks or dwell time.
  2. [§4.2.3, §4.2.4] λ=0.4 is a free parameter of the golden-list recipe, yet no sensitivity analysis is reported. Please show how NDCG@100 and the ablation conclusions vary with λ, and with the number of adjacent swaps used to create ranking pairs.
  3. [Figure 4, §5.2] The text says ranking-consistency metrics (WAS, PWKT, WMRD, DPA) 'remain essentially unchanged,' but no numbers are reported; please provide a table or at least the ranges.
  4. [§5.5.1] The LLM-as-judge study omits the judge model, the prompt template, and any consistency or self-bias checks; please add these details.
  5. [§4.6] The 80 ms latency claim is stated without specifying the hardware, batch size, and list-length conditions beyond K=10; please provide enough detail for reproducibility.
  6. [References] The reference list mixes arXiv preprints and venue papers without consistent citation keys for the LLM-related methods; please normalize per the target venue style.

Circularity Check

2 steps flagged · score 4.0 of 10

Partial circularity: supporting offline metrics mirror the hand-built golden-list training recipe, while the headline NDCG@100 and online A/B results remain independent.

  1. self definitional [§5.1.2 (Metrics), with §4.2.3–§4.2.4]
    "We evaluate four families of metrics that mirror the four preference-pair families used to train the reward."

    The four preference-pair families are generated by corrupting the hand-built golden list σ_gt along relevance, ranking, diversity, and redundancy axes (§4.2.3–§4.2.4). Both reward heads are trained on these pairs (§4.4.1–§4.4.2). The evaluation then measures ranking consistency (WAS, PWKT, WMRD, DPA), diversity (ILD), and redundancy (category entropy) against the same σ_gt-derived notions. Thus improvements on those metric families reflect how well the model reproduces the authors' own golden-list recipe rather than an external standard of page quality. The headline NDCG@100 and the online A/B test are outside this loop, so the circularity is partial.

  2. fitted input called prediction [§4.4.2 (Fine-grained reward), with §4.2.4 (Ranking pairs)]
    "To supervise such fine details, we define the item-level feedback set F = {(u, i_k, i'_k) | ΔE(u, i_k → i'_k) ≠ 0}, where ΔE(u, i_k → i'_k) is the change in user engagement caused by swapping item i_k with i'_k in the page of user u."

    The item-level reward is claimed to 'capture engagement-sensitive position swaps,' but the training objective L_fine is applied to Dreward, whose ranking pairs are constructed as 'a small number of adjacent swaps inside σ_gt_u' (§4.2.4). No measured click, purchase, dwell-time, or other engagement delta is used to define ΔE; the set F is never operationalized. The 'engagement-sensitive' label is therefore a relabeling of the rating-based golden-list ordering heuristic, and any claim that the fine head captures real engagement is not derived from engagement data.

full rationale

PageLLM's core pipeline is not circular in the strongest sense: the headline NDCG@100 is computed against held-out positive interactions (ratings), the online A/B test is an external business-metric check, and PPO improves over the SFT baseline that already imitates the golden list, so the NDCG gains cannot be reduced by construction to the reward's training labels. The self-citation [31] is incidental and not load-bearing. However, the paper's own §5.1.2 states that its ranking-consistency, diversity, and redundancy metrics 'mirror the four preference-pair families used to train the reward'; since those preference pairs are all generated by corrupting the hand-built golden list (§4.2.3, §4.2.4), the supporting metrics partially measure fit to the authors' own label recipe. Separately, the item-level reward is described as 'engagement-sensitive' even though its supervision comes from synthetic adjacent swaps of the rating-based golden list, with no measured engagement delta entering L_fine. These are real self-referential elements, but they do not force the main NDCG/online results.

Assumptions & free parameters 4 free parameters · 4 assumptions · 1 invented entities

The central claim rests on a hand-built golden list and synthetic preference pairs. The main free parameters are the category cap λ, the rating binarization threshold, and PPO hyperparameters. The invented 'engagement delta' is a conceptual placeholder that is never grounded in real engagement measurements.

free parameters (4)
  • lambda (category cap) = 0.4
    Hand-set in golden list construction (§4.2.3); caps the maximum fraction of items from any one category. Directly shapes the supervised target and therefore the preference pairs and reward.
  • rating threshold = r_ui > 3
    Binarizes explicit ratings into positive interactions (§4.2.2). Determines which items enter the golden list and the relevance labels used in evaluation.
  • PPO KL coefficient beta = 0.1
    Set in §5.1.4 for PPO fine-tuning; balances reward optimization against the reference policy.
  • number of adjacent swaps for ranking pairs = unspecified
    Ranking pairs are generated by 'a small number of adjacent swaps' (§4.2.4); the exact number and selection procedure are not stated.
assumptions (4)
  • domain assumption Explicit ratings binarized at r_ui > 3 are a valid proxy for user preference.
    Used to define positive interactions I+_u and the golden list (§4.2.2).
  • ad hoc to paper The constructed golden list (rating order, popularity tie-break, category cap λ=0.4) represents optimal whole-page quality.
    Introduced in §4.2.3; all preference pairs are corruptions of this list, so the reward model inherits this definition of quality.
  • domain assumption The four negative-list families isolate independent failure modes of page quality.
    Assumed in §4.2.4; no analysis is provided to show the corruptions do not overlap or miss other failure modes.
  • standard math Bradley-Terry pairwise preference modeling and PPO with KL penalty are appropriate for optimizing the combined reward.
    Standard RLHF machinery, cited [25, 28].
invented entities (1)
  • Engagement delta ΔE(u, i_k -> i'_k)
    purpose: Conceptual quantity intended to justify the item-level reward as capturing engagement-sensitive position swaps.
    Defined in Eq. 12 but never computed from logged engagement data; the item-level reward is trained on synthetic adjacent-swap preference pairs, so the connection to real engagement is asserted, not measured.

how reviews work

0 comments
Cite this review

Pith. "Pith review of PageLLM: A Multi-Grained Reward Framework for Whole-Page Optimization with Large Language Models." pith.science (2026). https://pith.science/paper/ZB52IGIN

@misc{pith2026250609084,
  author       = {Pith},
  title        = {Pith review of: PageLLM: A Multi-Grained Reward Framework for Whole-Page Optimization with Large Language Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZB52IGIN}},
  note         = {Machine review of arXiv:2506.09084}
}
read the original abstract

Whole-page optimization (WPO) decides how search and recommendation results are surfaced to users, and large language models (LLMs) open a new route to it by treating page generation as sequence generation. Adapting LLMs to web-scale WPO, however, remains bottlenecked by the need for costly human annotations and by the mismatched granularity between page-level coherence and item-level placement. In this work we show that these two challenges are coupled: implicit user feedback alone suffices for alignment, provided the reward signal is decoupled into two complementary granularities. We propose PageLLM, a reward-based fine-tuning framework that (i) turns implicit feedback into four contrastive preference-pair families covering relevance, ranking, diversity, and redundancy, (ii) learns a coarse page-level reward and a fine item-level reward that captures engagement-sensitive position swaps, and (iii) combines both rewards in PPO-based RLHF over a pre-trained LLM. Extensive experiments on seven Amazon categories against eleven baselines show that neither reward alone is sufficient -- dropping the page-level or item-level signal reduces NDCG@100 by 17.8% and 15.2% respectively, whereas the joint reward improves NDCG@100 by up to 46.8%. Deployed in a 10M-user online A/B test, PageLLM raises GMV by 0.44% and click-through rate by 0.14%, confirming that multi-grained rewards from implicit feedback scale to production WPO. Code and data are available at an anonymized repository.

Figures

Figures reproduced from arXiv: 2506.09084 by the authors.

Figure 1
Figure 1. Different ranking strategies yield different pages [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Overview of PageLLM. Implicit user feedback is turned into a golden ranked list and four families of preference [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Prompt template for fine-tuning PageLLM: a user [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Relative gain of PageLLM over its no-reward variant on seven Amazon categories. The multi-grained reward delivers [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Ablation of PageLLM’s reward heads on AM-Toys. Removing either head degrades every metric, showing that page [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 6
Figure 6. Figure 6: LLM-as-judge pairwise win-rate of PageLLM against [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]
Figure 7
Figure 7. Figure 7: Performance under a 50% cold-start simulation on [PITH_FULL_IMAGE:figures/full_fig_p009_7.png]
Figure 8
Figure 8. Figure 8: Performance vs. cost trade-off between GPT-2 and [PITH_FULL_IMAGE:figures/full_fig_p010_8.png]
Figure 9
Figure 9. Figure 9: A representative prediction trace. The generated [PITH_FULL_IMAGE:figures/full_fig_p010_9.png]

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

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

  1. Teaching Time Series to See and Speak: Forecasting with Aligned Visual and Textual Perspectives

    cs.LG 2025-06 reject novelty 5.0 of 10

    TimesCLIP aligns image-based and text-based views of the same time series via contrastive learning to improve forecasting accuracy on several benchmarks, but the full multimodal model is not used on two of the six lon...

Reference graph

Works this paper leans on

51 extracted references · 19 canonical work pages · cited by 1 Pith paper

  1. [1]

    Haoyue Bai, Min Hou, Le Wu, Yonghui Yang, Kun Zhang, Richang Hong, and Meng Wang. 2023. Gorec: a generative cold-start recommendation framework. In Proceedings of the 31st ACM international conference on multimedia. 1004–1012

    work page 2023

  2. [2]

    Haoyue Bai, Min Hou, Le Wu, Yonghui Yang, Kun Zhang, Richang Hong, and Meng Wang. 2024. Unified Representation Learning for Discrete Attribute En- hanced Completely Cold-Start Recommendation.IEEE Transactions on Big Data (2024)

    work page 2024

  3. [3]

    Haoyue Bai, Le Wu, Min Hou, Miaomiao Cai, Zhuangzhuang He, Yuyang Zhou, Richang Hong, and Meng Wang. 2024. Multimodality invariant learning for multimedia-based new item recommendation. InProceedings of the 47th Inter- national ACM SIGIR Conference on Research and Development in Information Retrieval. 677–686

    2024

  4. [4]

    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

    2023

  5. [5]

    Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, et al. 2020. Language models are few-shot learners.Advances in neural information processing systems33 (2020), 1877–1901

    2020

  6. [6]

    Miaomiao Cai, Min Hou, Lei Chen, Le Wu, Haoyue Bai, Yong Li, and Meng Wang

    work page

  7. [7]

    Zeyu Cui, Jianxin Ma, Chang Zhou, Jingren Zhou, and Hongxia Yang. 2022. M6-rec: Generative pretrained language models are open-ended recommender systems.arXiv preprint arXiv:2205.08084(2022)

    arXiv 2022

  8. [8]

    Jacob Devlin. 2018. Bert: Pre-training of deep bidirectional transformers for language understanding.arXiv preprint arXiv:1810.04805(2018)

    arXiv 2018

Show all 51 references
  1. [9]

    Dario Di Palma. 2023. Retrieval-augmented recommender system: Enhancing recommender systems with large language models. InProceedings of the 17th ACM Conference on Recommender Systems. 1369–1373

    2023

  2. [10]

    Weicong Ding, Dinesh Govindaraj, and SVN Vishwanathan. 2019. Whole page optimization with global constraints. InProceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. 3153–3161

    2019

  3. [11]

    Luke Friedman, Sameer Ahuja, David Allen, Zhenning Tan, Hakim Sidahmed, Changbo Long, Jun Xie, Gabriel Schubiner, Ajay Patel, Harsh Lara, et al. 2023. Leveraging large language models in conversational recommender systems.arXiv preprint arXiv:2305.07961(2023)

    2023 arXiv

  4. [12]

    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

    2022

  5. [13]

    Zhenhuan Gong et al. 2013. Multi-level Data Layout Optimization for Heteroge- neous Access Patterns. (2013)

    2013

  6. [14]

    Muhammad Usman Hadi, Rizwan Qureshi, Abbas Shah, Muhammad Irfan, Anas Zafar, Muhammad Bilal Shaikh, Naveed Akhtar, Jia Wu, Seyedali Mirjalili, et al

  7. [15]

    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

    2020

  8. [16]

    Zhuangzhuang He, Yifan Wang, Yonghui Yang, Peijie Sun, Le Wu, Haoyue Bai, Jinqi Gong, Richang Hong, and Min Zhang. 2024. Double correction frame- work for denoising recommendation. InProceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining. 1062–1072

    2024

  9. [17]

    Balázs Hidasi, Alexandros Karatzoglou, Linas Baltrunas, and Domonkos Tikk

  10. [18]

    Yupeng Hou, Shanlei Mu, Wayne Xin Zhao, Yaliang Li, Bolin Ding, and Ji-Rong Wen. 2022. Towards universal sequence representation learning for recommender systems. InProceedings of the 28th ACM SIGKDD Conference on Knowledge Dis- covery and Data Mining. 585–593

    2022

  11. [19]

    Wang-Cheng Kang and Julian McAuley. 2018. Self-attentive sequential recom- mendation. In2018 IEEE international conference on data mining (ICDM). IEEE, 197–206

    2018

  12. [20]

    Wang-Cheng Kang, Jianmo Ni, Nikhil Mehta, Maheswaran Sathiamoorthy, Lichan Hong, Ed Chi, and Derek Zhiyuan Cheng. 2023. Do llms understand user prefer- ences? evaluating llms on user rating prediction.arXiv preprint arXiv:2305.06474 (2023)

    2023 arXiv

  13. [21]

    Xinyi Li, Yongfeng Zhang, and Edward C Malthouse. 2023. Pbnr: Prompt-based news recommender system.arXiv preprint arXiv:2304.07862(2023)

    2023 arXiv

  14. [22]

    Dawen Liang, Rahul G Krishnan, Matthew D Hoffman, and Tony Jebara. 2018. Variational autoencoders for collaborative filtering. InProceedings of the 2018 world wide web conference. 689–698

    2018

  15. [23]

    Xinyu Lin, Wenjie Wang, Yongqi Li, Shuo Yang, Fuli Feng, Yinwei Wei, and Tat- Seng Chua. 2024. Data-efficient Fine-tuning for LLM-based Recommendation. InProceedings of the 47th International ACM SIGIR Conference on Research and Development in Information Retrieval. 365–374

    2024

  16. [24]

    Julian McAuley and Alex Yang. 2016. Addressing complex and subjective product- related queries with customer reviews. InProceedings of the 25th International Conference on World Wide Web. 625–635

    2016

  17. [25]

    Long Ouyang, Jeffrey Wu, Xu Jiang, Diogo Almeida, Carroll Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, et al. 2022. Training language models to follow instructions with human feedback. InAd- vances in Neural Information Processing Systems

    2022

  18. [26]

    Rafael Rafailov, Archit Sharma, Eric Mitchell, Christopher D Manning, Stefano Ermon, and Chelsea Finn. 2023. Direct preference optimization: Your language model is secretly a reward model. InAdvances in Neural Information Processing Systems

    2023

  19. [27]

    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 on Web Conference 2024. 3464– 3475

    2024

  20. [28]

    John Schulman, Filip Wolski, Prafulla Dhariwal, Alec Radford, and Oleg Klimov

  21. [29]

    Fei Sun, Jun Liu, Jian Wu, Changhua Pei, Xiao Lin, Wenwu Ou, and Peng Jiang

  22. [30]

    Sahil Verma, Ashudeep Singh, Varich Boonsanong, John P Dickerson, and Chirag Shah. 2023. RecRec: Algorithmic Recourse for Recommender Systems. InPro- ceedings of the 32nd ACM International Conference on Information and Knowledge Management. 4325–4329

    2023

  23. [31]

    Xinyuan Wang, Liang Wu, Liangjie Hong, Hao Liu, and Yanjie Fu. 2024. LLM- Enhanced User-Item Interactions: Leveraging Edge Information for Optimized Recommendations.arXiv preprint arXiv:2402.09617(2024)

    2024 arXiv

  24. [32]

    Yancheng Wang, Ziyan Jiang, Zheng Chen, Fan Yang, Yingxue Zhou, Eunah Cho, Xing Fan, Xiaojiang Huang, Yanbin Lu, and Yingzhen Yang. 2023. Rec- mind: Large language model powered agent for recommendation.arXiv preprint arXiv:2308.14296(2023)

    2023 arXiv

  25. [33]

    Yue Wang, Dawei Yin, Luo Jie, Pengyuan Wang, Makoto Yamada, Yi Chang, and Qiaozhu Mei. 2016. Beyond ranking: Optimizing whole-page presentation. In Proceedings of the Ninth ACM International Conference on Web Search and Data Mining. 103–112

    2016

  26. [34]

    Le Wu, Xiangnan He, Xiang Wang, Kun Zhang, and Meng Wang. 2022. A survey on accuracy-oriented neural recommendation: From collaborative filtering to information-rich recommendation.IEEE Transactions on Knowledge and Data Engineering35, 5 (2022), 4425–4445

    2022

  27. [35]

    Dehong Xu, Liang Qiu, Minseok Kim, Faisal Ladhak, and Jaeyoung Do. 2024. Aligning Large Language Models via Fine-grained Supervision.arXiv preprint arXiv:2406.02756(2024)

    2024 arXiv

  28. [36]

    Lanling Xu, Junjie Zhang, Bingqian Li, Jinpeng Wang, Mingchen Cai, Wayne Xin Zhao, and Ji-Rong Wen. 2024. Prompting large language models for recommender systems: A comprehensive framework and empirical analysis.arXiv preprint arXiv:2401.04997(2024)

    2024 arXiv

  29. [37]

    Jing Yao, Wei Xu, Jianxun Lian, Xiting Wang, Xiaoyuan Yi, and Xing Xie. 2023. Knowledge plugins: Enhancing large language models for domain-specific rec- ommendations.arXiv preprint arXiv:2311.10779(2023)

    2023 arXiv

  30. [38]

    Gangman Yi, Donghoon Kim, and Neil Yen. 2022. Computational optimization and applications for heterogeneous multimedia data. (2022)

    2022

  31. [39]

    Zhenrui Yue, Sara Rabhi, Gabriel de Souza Pereira Moreira, Dong Wang, and Even Oldridge. 2023. LlamaRec: Two-stage recommendation using large language models for ranking.arXiv preprint arXiv:2311.02089(2023)

    2023 arXiv

  32. [40]

    Jiaqi Zhai, Lucy Liao, Xing Liu, Yueming Wang, Rui Li, Xuan Cao, Leon Gao, Zhao- jie Gong, Fangda Gu, Michael He, et al. 2024. Actions speak louder than words: Trillion-parameter sequential transducers for generative recommendations.arXiv preprint arXiv:2402.17152(2024)

    2024 arXiv

  33. [41]

    Junjie Zhang, Ruobing Xie, Yupeng Hou, Wayne Xin Zhao, Leyu Lin, and Ji-Rong Wen. 2023. Recommendation as instruction following: A large language model empowered recommendation approach.arXiv preprint arXiv:2305.07001(2023)

    2023 arXiv

  34. [42]

    Tingting Zhang, Pengpeng Zhao, Yanchi Liu, Victor S Sheng, Jiajie Xu, Deqing Wang, Guanfeng Liu, Xiaofang Zhou, et al . 2019. Feature-level Deeper Self- Attention Network for Sequential Recommendation.. InIJCAI. 4320–4326

    2019

  35. [43]

    Yue Zhang, Yafu Li, Leyang Cui, Deng Cai, Lemao Liu, Tingchen Fu, Xinting Huang, Enbo Zhao, Yu Zhang, Yulong Chen, et al . 2023. Siren’s song in the AI ocean: a survey on hallucination in large language models.arXiv preprint arXiv:2309.01219(2023)

    2023 arXiv

  36. [44]

    Wayne Xin Zhao, Kun Zhou, Junyi Li, Tianyi Tang, Xiaolei Wang, Yupeng Hou, Yingqian Min, Beichen Zhang, Junjie Zhang, Zican Dong, et al. 2023. A survey of large language models.arXiv preprint arXiv:2303.18223(2023). Conference’17, July 2017, Washington, DC, USA Anonymous et al

    2023 arXiv

  37. [45]

    Kun Zhou, Hui Wang, Wayne Xin Zhao, Yutao Zhu, Sirui Wang, Fuzheng Zhang, Zhongyuan Wang, and Ji-Rong Wen. 2020. S3-rec: Self-supervised learning for se- quential recommendation with mutual information maximization. InProceedings of the 29th ACM international conference on inf...

    2020

  38. [46]

    Yaochen Zhu and Zhenzhong Chen. 2022. Mutually-regularized dual collaborative variational auto-encoder for recommendation systems. InProceedings of The ACM Web Conference 2022. 2379–2387

    2022

  39. [2015]

    Session-based recommendations with recurrent neural networks.arXiv preprint arXiv:1511.06939(2015)

    2015 arXiv

  40. [2017]

    Proximal policy optimization algorithms.arXiv preprint arXiv:1707.06347 (2017)

    2017 arXiv

  41. [2019]

    InProceedings of the 28th ACM international conference on information and knowledge management

    BERT4Rec: Sequential recommendation with bidirectional encoder rep- resentations from transformer. InProceedings of the 28th ACM international conference on information and knowledge management. 1441–1450

  42. [2023]

    A survey on large language models: Applications, challenges, limitations, and practical usage.Authorea Preprints(2023)

    2023

  43. [2024]

    Mitigating Recommendation Biases via Group-Alignment and Global- Uniformity in Representation Learning.ACM Transactions on Intelligent Systems and Technology(2024)

    2024

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.