Pith. sign in

REVIEW 4 major objections 5 minor 13 cited by

Towards Large-scale Generative Ranking

T0 review · 4 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read Generative ranking beats the production model at equal compute.

desk verdict Real industrial evidence that generative ranking can work online, but the resource-equivalence claim is asserted rather than measured and the architecture conclusion rests on near-threshold offline deltas. read the letter →

arxiv 2505.04180 v2 pith:OR7RCH3W submitted 2025-05-07 cs.IR

Yanhua Huang , Yuqi Chen , Xiong Cao , Rui Yang , Mingliang Qi , Yinghao Zhu , Qingchang Han , Yaowei Liu
show 12 more authors
Zhaoyu Liu Xuefeng Yao Yuting Jia Leilei Ma Yinqi Zhang Taoyu Zhu Liujie Zhang Lei Chen Weihang Chen Min Zhu Ruiwen Xu Lei Zhang
This is my paper · ORCID
classification cs.IR
keywords generativerankingrecommendationaction-orientedorganizationpositionandtimebiasesALiBirecommendersystemsonlineA/Btestingindustrialdeployment
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 asks what actually makes generative ranking work in a large-scale industrial recommender, and whether it can be deployed without blowing up compute. It argues that the gains come from the generative architecture itself, not from how training samples are grouped or ordered. To make that architecture practical, the paper introduces GenRank, which treats items as positional context and predicts the user's actions on candidate items, along with lightweight position and time biases. Online experiments on a social-media explore feed serving tens of millions of users show GenRank improving time spent, reads, engagements, and seven-day retention at roughly the same compute as the production ranking model, with a more than 25 percent better P99 response time.

What carries the argument

The central object is GenRank's action-oriented organization, which treats items as positional indicators and the action as the fundamental generative unit: each input token sums the item embedding and action embedding, with a mask embedding for candidate items, so the sequence length is halved relative to interleaved item-action designs. The causal transformer decoder with a candidate mask generates actions autoregressively and prevents information leakage between candidates. Three additive embeddings (position, request index, pre-request time gap) plus the parameter-free ALiBi distance bias replace the learned relative attention bias, lowering attention costs by about 75 percent and projection costs by 50 percent, and reducing bias-related I/O from quadratic to linear.

What would settle it

Run GenRank and the baseline with multiple seeds and compute confidence intervals for the AUC differences; if the +0.0009 and +0.0006 differences are statistically indistinguishable from zero at the stated 0.0010 significance threshold, the claim that the new biases and action-oriented organization preserve accuracy while adding efficiency is not supported.

Watch

Extended reading notes

Core claim

The central claim is that the effectiveness of generative ranking in large-scale industrial settings comes primarily from the generative architecture, not from the training paradigm. The paper supports this by ablating an existing generative sequential transducer: computing the loss at historical behavior positions reduces AUC by more than 0.0100, and replacing the causal mask with a fully visible mask at historical positions reduces AUC by more than 0.0015, a drop that grows with model size; in contrast, reordering grouped training samples into a point-wise order causes only a slight decrease. The paper then introduces GenRank, an action-oriented generative architecture in which each position is one item-action pair, candidate items are represented with a mask action embedding, and a causal transformer with a candidate mask predicts the user's action on each candidate. GenRank adds three additive embeddings (position, request index, pre-request time gap) and the parameter-free ALiBi relative bias, halving attention sequence length and cutting bias I/O from quadratic to linear. Offline, GenRank cuts training time by 94.8 percent relative to the baseline with a slight AUC improvement; online, it improves time spent by 0.3345 percent, reads by 0.6325 percent, engagements by 1.2474 percent, and seven-day retention by 0.1481 percent at comparable overall resources.

Load-bearing premise

The load-bearing premise is that an absolute difference of 0.0010 in the ranking-quality metric AUC on the main tasks counts as practically significant, since the ablation differences separating GenRank's components (-0.0003, +0.0009, +0.0006) are near or below that threshold; if those differences are not meaningful, the efficiency argument still holds but the accuracy-preservation claim does not.

Editorial extensions

If this is right

  • GenRank's 94.8 percent training speedup at equal or slightly better AUC means generative ranking can be trained at production scale without sacrificing effectiveness.
  • The result that content embeddings more than double their AUC contribution under the generative paradigm implies that pre-trained multimodal representations are better exploited when the ranking model generates actions.
  • The minimal feature engineering and KV-cache scalability of GenRank point toward a future where one generative model could serve both pre-ranking and ranking stages.
  • The online gains on time spent, reads, engagements, and seven-day retention show the improvement is not confined to a single engagement metric.

Reading between the lines

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

  • If the architecture rather than the training paradigm is the driver, then a model trained on point-wise ordered logs with the same GenRank architecture should retain most of the accuracy gain; this is a cheap replication test for other teams.
  • Because GenRank's action-oriented representation is agnostic to action type, the same design should extend to predicting follow, share, save, and purchase actions without architectural change.
  • The more than 25 percent improvement in P99 response time suggests the latency budget could be reinvested in test-time scaling (more steps or larger candidate sets) rather than only in cost savings.
  • The reported cold-start gains, if they hold, imply that content embeddings are doing more work under generative ranking; a targeted online experiment isolating cold-start traffic would quantify that mechanism.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper studies generative ranking at the ranking stage of Xiaohongshu's Explore Feed. It makes three main claims: (i) based on ablations built around HSTU, the effectiveness of generative ranking comes primarily from the generative architecture rather than from the training paradigm or sample organization; (ii) the proposed GenRank architecture, which uses an action-oriented sequence organization and new position, request-index, and pre-request time embeddings with ALiBi, preserves accuracy while speeding up training relative to HSTU by 94.8%; and (iii) a 15-day online A/B experiment on 10% user buckets shows that GenRank improves time spent, reads, engagements, and LT7 relative to the production ranker, with 'nearly equivalent computational resources.' The paper reports no confidence intervals or p-values for the online metrics, provides only qualitative support for the resource-equivalence claim, and reports offline AUC differences that are at or below the paper's own significance threshold.

Significance. If the claims hold, this is a valuable industrial data point: it suggests that generative ranking can be deployed at feed scale with acceptable cost, and it identifies the generative architecture as the main driver of effectiveness, which is useful guidance for future designs. The large-scale online A/B experiment on tens of millions of users and the large-scale offline experiments are genuine strengths. However, the paper's practical-value argument rests on two pillars that are currently under-evidenced: the offline AUC differences are near or below the stated 0.0010 significance threshold, and the 'nearly equivalent computational resources' claim is not quantified. The architecture-versus-paradigm conclusion is also based on a single generative baseline, HSTU, so the generality of the conclusion remains open. These issues are fixable with additional reporting and analysis, but they are central to the paper's headline claims.

major comments (4)
  1. [Section 6, 'In terms of overhead' paragraph] The central feasibility claim that GenRank achieves its online gains with 'nearly equivalent computational resources' is not quantified. The text states that GenRank incurs higher training costs but lower inference and storage costs, and reports a P99 latency improvement of over 25%, but it gives no training GPU-hours, cost per request, storage footprint, or total-cost comparison. Moreover, Table 1 measures training speed-ups relative to HSTU, which is not the production baseline used in the online A/B experiment, so Table 1 cannot substantiate equivalence to the production system. Since this claim underpins contribution 3 and the paper's practical-value argument, the authors should report concrete cost numbers or substantially temper the claim.
  2. [Section 2 and Table 1] The paper sets an absolute AUC increase of 0.0010 on main tasks as the threshold for practical significance. The ablation results in Table 1 are -0.0003 for the action-oriented organization, +0.0009 for the proposed position and time biases, and +0.0006 for GenRank overall; these are at or below the stated threshold, and no confidence intervals or repeated-run variation are reported. Consequently, the claims that the action-oriented organization preserves accuracy and that the new biases improve accuracy are not established by the reported numbers. The architecture-versus-paradigm conclusion also rests on a single baseline, HSTU, with no comparison against the production architecture under identical training conditions.
  3. [Section 3.1, 'Key Mechanisms in Generative Paradigm'] The two experiments that support the central claim that the auto-regressive manner is critical are described only qualitatively: the text reports an 'AUC drop of more than 0.0100' when loss is computed at historical positions and an 'AUC drop of more than 0.0015' when the causal mask is replaced with a fully visible mask at historical positions. No table, exact metrics, or hyperparameter configurations are provided for these experiments. Because this is the core evidence for attributing effectiveness to the architecture, the authors should report the full numbers, ideally with variation across seeds, and specify the model sizes and settings used for the mask and loss-position variants.
  4. [Section 6, Table 2] The online A/B result table reports only mean percentage improvements over a 15-day period, with no confidence intervals, p-values, standard errors, or day-level data. The word 'significant' in the abstract and Section 6 is therefore not statistically supported by the reported results. The authors should report the distribution of daily lifts or provide confidence intervals and p-values for each of the four metrics, and state whether any multiple-comparison correction was applied.
minor comments (5)
  1. [Section 6] The abbreviation 'GAUC' is used without definition; please define it or specify its computation.
  2. [Section 4.2] The bucket boundaries for the pre-request time embedding and the maximum sequence length are not specified; these are free parameters that affect reproducibility.
  3. [Abstract and Section 1] The abstract and introduction refer to 'theoretical analyses,' but Section 3 presents empirical experiments; either add a formal theoretical statement or soften the wording.
  4. [Section 4.1, Equation (1)] Equation (1) defines e^{(p,t)}_i, but the text uses E_{pe,i}, E_{ri,i}, and E_{rt,i} without defining the superscript; please align the notation.
  5. [Section 3.2] The claim that content embeddings yield over twice the AUC improvement under the generative paradigm is made without giving the underlying AUC values; please include the numbers.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: empirical ablations against external baselines carry the claims; all self-citations are peripheral.

full rationale

The paper's central claims rest on direct experiments, not on fitted inputs being renamed as predictions. Section 3 tests the auto-regressive mechanism by computing loss at historical positions and by swapping the causal mask for a fully visible mask; both are controlled interventions reported as AUC changes. The conclusion that architecture, not sample organization, drives effectiveness follows from comparing grouped point-wise training with the standard generative organization, again a direct ablation. Table 1 reports speed-ups from halving the attention sequence length and from replacing quadratic-attention I/O with linear embeddings plus ALiBi; these are consequences of the stated architectural design, not of optimizing against the evaluation metric. Online A/B results compare GenRank against the production model, an external benchmark. The authors' own references ([6] on neural statistics and [7] on sliding spectrum decomposition) are cited only for peripheral feature-engineering and content-embedding modules, and are not load-bearing for the generative-architecture conclusion; there is no author-imported uniqueness theorem or ansatz-by-citation. The paper does state that an absolute AUC gain of 0.0010 is considered significant, and it does leave the 'nearly equivalent computational resources' claim unquantified (higher training cost, lower inference/storage cost, P99 improvement over 25% without totals). These are substantive evidential limitations and correctness-risk concerns, but they are not circularity: no equation reduces to its own input, and no fitted parameter is presented as an independent prediction. Accordingly, no circular steps are identified.

Assumptions & free parameters 3 free parameters · 5 assumptions · 4 invented entities

The central claim relies on an autoregressive sequence-modeling assumption, a proprietary data pipeline, and several hand-chosen hyperparameters and embeddings. No code or external benchmarks are provided, so the new architecture components are supported only by internal experiments.

free parameters (3)
  • Maximum sequence length 480 = 480
    Set by hand in Section 3; attention cost and speed-up figures depend on this limit.
  • Model size hyperparameters (blocks 3, heads 8, hidden 768) = 3, 8, 768
    Baseline and GenRank share these settings; no sensitivity analysis is reported in Section 3.
  • Time bucket boundaries for pre-request time embedding = not specified
    Defined in Section 4.2 with bucket() but the boundaries are not given, making the representation under-specified.
assumptions (5)
  • domain assumption User actions can be modeled autoregressively conditioned on item context.
    The model factorizes p(a_k | x_1, a_1, ..., x_k) in Section 4.1; this is the core modeling premise and is not derived.
  • domain assumption An AUC increase of at least 0.0010 on main tasks indicates practical significance.
    Stated in Section 2; the ablation diffs in Table 1 are near or below this threshold, so the architecture conclusions rely on this judgment.
  • domain assumption HSTU is a fair, representative baseline for generative ranking.
    Section 3 uses HSTU as the only generative baseline; no tuning budget or implementation validation is reported.
  • domain assumption Random 10% user split gives unbiased online estimates.
    Section 6 defines control and treatment groups but reports no covariate balance or interference checks.
  • domain assumption Frozen content embeddings contribute transferable world knowledge.
    Section 3.2 attributes larger content-embedding gains to architectural consistency, presupposing the embeddings already encode useful item semantics.
invented entities (4)
  • Mask action embedding M
    purpose: Represents unknown future actions for candidate items so the model predicts actions only (Section 4.1, Figure 2b).
    Supported only by internal ablations; no external benchmark or independent handle.
  • Request index embedding E_ri
    purpose: Encodes which request an item belongs to, grouping items from the same user request (Section 4.2).
    Shown only through the internal speed-up table; no external validation.
  • Pre-request time embedding E_rt
    purpose: Encodes bucketed time since the previous request to model user activity level (Section 4.2).
    Bucket boundaries are unspecified; no external evidence.
  • Action-oriented sequence organization
    purpose: Halves token sequence length by treating items as position context and actions as prediction units (Section 4.1).
    Speed-up measured only on proprietary data against HSTU.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Towards Large-scale Generative Ranking." pith.science (2026). https://pith.science/paper/OR7RCH3W

@misc{pith2026250504180,
  author       = {Pith},
  title        = {Pith review of: Towards Large-scale Generative Ranking},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OR7RCH3W}},
  note         = {Machine review of arXiv:2505.04180}
}
read the original abstract

Generative recommendation has recently emerged as a promising paradigm in information retrieval. However, generative ranking systems are still understudied, particularly with respect to their effectiveness and feasibility in large-scale industrial settings. This paper investigates this topic at the ranking stage of Xiaohongshu's Explore Feed, a recommender system that serves hundreds of millions of users. Specifically, we first examine how generative ranking outperforms current industrial recommenders. Through theoretical and empirical analyses, we find that the primary improvement in effectiveness stems from the generative architecture, rather than the training paradigm. To facilitate efficient deployment of generative ranking, we introduce GenRank, a novel generative architecture for ranking. We validate the effectiveness and efficiency of our solution through online A/B experiments. The results show that GenRank achieves significant improvements in user satisfaction with nearly equivalent computational resources compared to the existing production system.

Figures

Figures reproduced from arXiv: 2505.04180 by the authors.

Figure 1
Figure 1. Left: A screenshot of Xiaohongshu’s Explore Feed product. Right: Illustration of cascade [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Model architecture of GenRank. Compared to existing approaches, e.g., HSTU [ [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Illustration of input representation and candidate mask. (a) The input representation of [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 13 Pith papers

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

  1. Knowledge-Geometry Decoupling: Refreshable Pretrained Transfer for Streaming Recommendation

    cs.IR 2026-08 conditional novelty 7.0 of 10

    KGD decouples a refreshable pretrained encoder from a task learner via read-only cross-attention and an orthogonal residual, improving streaming recommendation accuracy and surviving 90 days of distribution drift.

  2. Asymmetric Generative Recommendation via Kronecker Residual Bridge and Multi-Faceted Hierarchical Quantization

    cs.IR 2026-05 unverdicted novelty 7.0 of 10

    AsymRec decouples input and output representations in generative recommendation via multi-expert semantic projection and multi-faceted hierarchical quantization, outperforming prior models by 15.8% on average.

  3. Population-Level Generative Modeling for Ranking Data

    stat.ME 2026-08 conditional novelty 6.0 of 10

    Ranking generation is reduced to learning a latent preference distribution on a simplex, with flow matching used to sample new preference weights and the fitted Plackett-Luce model used to decode rankings.

  4. UniVA: Unified Value Alignment for Generative Recommendation in Online Advertising at Tencent

    cs.IR 2026-05 unverdicted novelty 6.0 of 10

    UniVA unifies value alignment in generative recommendation via a Commercial SID tokenizer, eCPM-aware RL decoder, and personalized beam search, reporting 37% offline Hit Rate gains and 1.5% online GMV lift on Tencent ...

  5. Deep Interest Mining for Intent-Enriched Semantic IDs in Multimodal Generative Recommendation

    cs.IR 2026-03 unverdicted novelty 6.0 of 10

    A new framework integrating deep interest mining, cross-modal semantic alignment, and quality-aware reinforcement learning generates higher-quality Semantic IDs and outperforms prior methods on recommendation benchmarks.

  6. UniSearch: Rethinking Search System with a Unified Generative Architecture

    cs.IR 2025-09 conditional novelty 6.0 of 10

    UniSearch replaces a cascaded recall and ranking pipeline with one generative model that jointly learns video semantic IDs and query-to-ID generation, improving offline and online search metrics at Kuaishou.

  7. Request-Only Optimization for Recommendation Systems

    cs.IR 2025-07 conditional novelty 6.0 of 10

    A request-level training data format eliminates duplicate user features, increasing storage efficiency and training throughput while enabling larger recommendation architectures.

  8. LLM-Based Generative Retrieval for Snapchat Content Recommendation

    cs.IR 2026-07 conditional novelty 5.0 of 10

    SnapLGR, a production LLM-based generative retrieval system for Snapchat short video, lifted View Time 0.37% and related engagement metrics in a 7-day A/B test, with offline ablation attributing most of the gain to de...

  9. TSGR: Taobao Search Generative Retrieval

    cs.IR 2026-07 conditional novelty 5.0 of 10

    TSGR makes Taobao search retrieval value-aware by ordering item IDs by clicks and query intent and re-ranking generated candidates with a value module, improving HR@1000 and GMV.

  10. Beyond Quality: Unlocking Diversity in Ad Headline Generation with Large Language Models

    cs.CL 2025-08 conditional novelty 5.0 of 10

    DIVER trains an LLM with synthetic data plus multi-objective RL to emit several diverse, style-tagged ad headlines in one pass, reporting online ADVV and CTR gains.

  11. Token Factory: Efficiently Integrating Diverse Signals into Large Recommendation Models

    cs.IR 2026-06 unverdicted novelty 4.0 of 10

    Token Factory transforms traditional signals into soft tokens for efficient integration and compression into Large Recommendation Models, avoiding prompt length explosion while enhancing performance.

  12. A Survey of Real-World Recommender Systems: Challenges, Constraints, and Industrial Perspectives

    cs.IR 2025-09 conditional novelty 4.0 of 10

    A survey of A/B-validated industrial recommender systems, split into transaction-oriented and content-oriented categories, with a discussion of the academia-industry gap.

  13. GR-LLMs: Recent Advances in Generative Recommendation Based on Large Language Models

    cs.IR 2025-07 unverdicted novelty 3.0 of 10

    A survey of LLM-based generative recommendation systems, covering application settings, training pipelines, industrial deployment challenges, and future directions.

Reference graph

Works this paper leans on

29 extracted references · 22 canonical work pages · cited by 13 Pith papers

  1. [1]

    Exploring Training and Inference Scaling Laws in Generative Retrieval

    Cai Hongru, Li Yongqi, Yuan Ruifeng, Wang Wenjie, Zhang Zhen, Li Wenjie, Chua Tat-Seng. Exploring Training and Inference Scaling Laws in Generative Retrieval // arXiv preprint arXiv:2503.18941. 2025

    work page Pith review arXiv 2025

  2. [2]

    Chang Jianxin, Zhang Chenbin, Fu Zhiyi, Zang Xiaoxue, Guan Lin, Lu Jing, Hui Yiqun, Leng Dewei, Niu Yanan, Song Yang, others . TWIN: TWo-stage interest network for lifelong user behavior modeling in CTR prediction at kuaishou // Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining. 2023. 3785–3794

    work page 2023

  3. [3]

    Chang Jianxin, Zhang Chenbin, Hui Yiqun, Leng Dewei, Niu Yanan, Song Yang, Gai Kun . Pepnet: Parameter and embedding personalized network for infusing with personalized prior information // Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining. 2023. 3795–3804

    work page 2023

  4. [4]

    Wide & deep learning for recommender systems // Proceedings of the 1st workshop on deep learning for recommender systems

    Cheng Heng-Tze, Koc Levent, Harmsen Jeremiah, Shaked Tal, Chandra Tushar, Aradhye Hrishi, Anderson Glen, Corrado Greg, Chai Wei, Ispir Mustafa, others . Wide & deep learning for recommender systems // Proceedings of the 1st workshop on deep learning for recommender systems. 2016. 7–10

    work page 2016

  5. [5]

    Scaling New Frontiers: Insights into Large Recommendation Models // arXiv preprint arXiv:2412.00714

    Guo Wei, Wang Hao, Zhang Luankang, Chin Jin Yao, Liu Zhongzhou, Cheng Kai, Pan Qiushi, Lee Yi Quan, Xue Wanqi, Shen Tingjia, others . Scaling New Frontiers: Insights into Large Recommendation Models // arXiv preprint arXiv:2412.00714. 2024

    arXiv 2024

  6. [6]

    Neural statistics for click-through rate prediction // Proceedings of the 45th International ACM SIGIR Conference on Research and Development in Information Retrieval

    Huang Yanhua, Wang Hangyu, Miao Yiyun, Xu Ruiwen, Zhang Lei, Zhang Weinan . Neural statistics for click-through rate prediction // Proceedings of the 45th International ACM SIGIR Conference on Research and Development in Information Retrieval. 2022. 1849–1853

    work page 2022

  7. [7]

    Sliding spectrum decomposition for diversified recommendation // Proceedings of the 27th ACM SIGKDD conference on knowledge discovery & data mining

    Huang Yanhua, Wang Weikun, Zhang Lei, Xu Ruiwen. Sliding spectrum decomposition for diversified recommendation // Proceedings of the 27th ACM SIGKDD conference on knowledge discovery & data mining. 2021. 3041–3049

    work page 2021

  8. [8]

    Self-attentive sequential recommendation // 2018 IEEE international conference on data mining (ICDM)

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

    work page 2018

Show all 29 references
  1. [9]

    Scaling laws for neural language models // arXiv preprint arXiv:2001.08361

    Kaplan Jared, McCandlish Sam, Henighan Tom, Brown Tom B, Chess Benjamin, Child Rewon, Gray Scott, Radford Alec, Wu Jeffrey, Amodei Dario. Scaling laws for neural language models // arXiv preprint arXiv:2001.08361. 2020

    2001 arXiv

  2. [10]

    Pi Qi, Zhou Guorui, Zhang Yujing, Wang Zhe, Ren Lejian, Fan Ying, Zhu Xiaoqiang, Gai Kun. Search-based user interest modeling with lifelong sequential behavior data for click-through rate prediction // Proceedings of the 29th ACM International Conference on Information & Knowl...

    2020

  3. [11]

    Train short, test long: Attention with linear biases enables input length extrapolation // arXiv preprint arXiv:2108.12409

    Press Ofir, Smith Noah A, Lewis Mike . Train short, test long: Attention with linear biases enables input length extrapolation // arXiv preprint arXiv:2108.12409. 2021

    2021 arXiv

  4. [12]

    Exploring the limits of transfer learning with a unified text-to-text transformer // Journal of machine learning research

    Raffel Colin, Shazeer Noam, Roberts Adam, Lee Katherine, Narang Sharan, Matena Michael, Zhou Yanqi, Li Wei, Liu Peter J . Exploring the limits of transfer learning with a unified text-to-text transformer // Journal of machine learning research. 2020. 21, 140. 1–67

    2020

  5. [13]

    Recommender systems with generative retrieval // Advances in Neural Information Processing Systems

    Rajput Shashank, Mehta Nikhil, Singh Anima, Hulikal Keshavan Raghunandan, Vu Trung, Heldt Lukasz, Hong Lichan, Tay Yi, Tran Vinh, Samost Jonah, others. Recommender systems with generative retrieval // Advances in Neural Information Processing Systems. 2023. 36. 10299–10315

    2023

  6. [14]

    Non-autoregressive generative models for reranking recommendation // Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining

    Ren Yuxin, Yang Qiya, Wu Yichun, Xu Wei, Wang Yalong, Zhang Zhiqiang. Non-autoregressive generative models for reranking recommendation // Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining. 2024. 5625–5634

    2024

  7. [15]

    Si Zihua, Guan Lin, Sun ZhongXiang, Zang Xiaoxue, Lu Jing, Hui Yiqun, Cao Xingchao, Yang Zeyu, Zheng Yichen, Leng Dewei, others. Twin v2: Scaling ultra-long user behavior sequence modeling for enhanced ctr prediction at kuaishou // Proceedings of the 33rd ACM International Con...

    2024

  8. [16]

    Sun Fei, Liu Jun, Wu Jian, Pei Changhua, Lin Xiao, Ou Wenwu, Jiang Peng . BERT4Rec: Sequential recommendation with bidirectional encoder representations from transformer // Pro- ceedings of the 28th ACM international conference on information and knowledge management

  9. [17]

    Progressive layered extraction (ple): A novel multi-task learning (mtl) model for personalized recommendations // Proceedings of the 14th ACM conference on recommender systems

    Tang Hongyan, Liu Junning, Zhao Ming, Gong Xudong. Progressive layered extraction (ple): A novel multi-task learning (mtl) model for personalized recommendations // Proceedings of the 14th ACM conference on recommender systems. 2020. 269–278

    2020

  10. [18]

    Learnable item tokenization for generative recommendation // Proceedings of the 33rd ACM International Conference on Information and Knowledge Management

    Wang Wenjie, Bao Honghui, Lin Xinyu, Zhang Jizhi, Li Yongqi, Feng Fuli, Ng See-Kiong, Chua Tat-Seng. Learnable item tokenization for generative recommendation // Proceedings of the 33rd ACM International Conference on Information and Knowledge Management. 2024. 2400–2409

    2024

  11. [19]

    EAGER: Two-Stream Generative Recommender with Behavior- Semantic Collaboration // Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining

    Wang Ye, Xun Jiahao, Hong Minjie, Zhu Jieming, Jin Tao, Lin Wang, Li Haoyuan, Li Linjun, Xia Yan, Zhao Zhou, others . EAGER: Two-Stream Generative Recommender with Behavior- Semantic Collaboration // Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data...

    2024

  12. [20]

    Content-Based Collaborative Generation for Recom- mender Systems // Proceedings of the 33rd ACM International Conference on Information and Knowledge Management

    Wang Yidan, Ren Zhaochun, Sun Weiwei, Yang Jiyuan, Liang Zhixiang, Chen Xin, Xie Ruobing, Yan Su, Zhang Xu, Ren Pengjie, others. Content-Based Collaborative Generation for Recom- mender Systems // Proceedings of the 33rd ACM International Conference on Information and Knowledg...

    2024

  13. [21]

    Enhancing Performance and Scalability of Large-Scale Recommendation Systems with Jagged Flash Attention // Proceedings of the 18th ACM Conference on Recommender Systems

    Xu Rengan, Yang Junjie, Xu Yifan, Li Hong, Liu Xing, Shankar Devashish, Zhang Haoci, Liu Meng, Li Boyang, Hu Yuxi, others . Enhancing Performance and Scalability of Large-Scale Recommendation Systems with Jagged Flash Attention // Proceedings of the 18th ACM Conference on Reco...

    2024

  14. [22]

    Sparse Meets Dense: Unified Generative Recommendations with Cascaded Sparse-Dense Representations // arXiv preprint arXiv:2503.02453

    Yang Yuhao, Ji Zhi, Li Zhaopeng, Li Yi, Mo Zhonglin, Ding Yue, Chen Kai, Zhang Zijian, Li Jie, Li Shuanglong, others . Sparse Meets Dense: Unified Generative Recommendations with Cascaded Sparse-Dense Representations // arXiv preprint arXiv:2503.02453. 2025

    2025 arXiv

  15. [23]

    Actions speak louder than words: Trillion-parameter sequential transducers for generative recommendations // arXiv preprint arXiv:2402.17152

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

    2024 arXiv

  16. [24]

    Scaling vision transformers // Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Zhai Xiaohua, Kolesnikov Alexander, Houlsby Neil, Beyer Lucas. Scaling vision transformers // Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2022. 12104–12113

    2022

  17. [25]

    Sigmoid loss for language image pre-training // Proceedings of the IEEE/CVF international conference on computer vision

    Zhai Xiaohua, Mustafa Basil, Kolesnikov Alexander, Beyer Lucas. Sigmoid loss for language image pre-training // Proceedings of the IEEE/CVF international conference on computer vision

  18. [26]

    Wukong: Towards a scaling law for large-scale recommendation // arXiv preprint arXiv:2403.02545

    Zhang Buyun, Luo Liang, Chen Yuxin, Nie Jade, Liu Xi, Guo Daifeng, Zhao Yanli, Li Shen, Hao Yuchen, Yao Yantao, others. Wukong: Towards a scaling law for large-scale recommendation // arXiv preprint arXiv:2403.02545. 2024

    2024 arXiv

  19. [27]

    Notellm-2: Multimodal large representation models for recommendation // arXiv preprint arXiv:2405.16789

    Zhang Chao, Zhang Haoxin, Wu Shiwei, Wu Di, Xu Tong, Zhao Xiangyu, Gao Yan, Hu Yao, Chen Enhong. Notellm-2: Multimodal large representation models for recommendation // arXiv preprint arXiv:2405.16789. 2024

    2024 arXiv

  20. [28]

    Towards understanding the overfitting phenomenon of deep click-through rate models // Proceedings of the 31st ACM international conference on information & knowledge management

    Zhang Zhao-Yu, Sheng Xiang-Rong, Zhang Yujing, Jiang Biye, Han Shuguang, Deng Hongbo, Zheng Bo. Towards understanding the overfitting phenomenon of deep click-through rate models // Proceedings of the 31st ACM international conference on information & knowledge management. 202...

    2022

  21. [29]

    Deep interest network for click-through rate prediction // Proceedings of the 24th ACM SIGKDD international conference on knowledge discovery & data mining

    Zhou Guorui, Zhu Xiaoqiang, Song Chenru, Fan Ying, Zhu Han, Ma Xiao, Yan Yanghui, Jin Junqi, Li Han, Gai Kun. Deep interest network for click-through rate prediction // Proceedings of the 24th ACM SIGKDD international conference on knowledge discovery & data mining

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.