Pith. sign in

REVIEW 3 major objections 5 minor 2 cited by

Request-Only Optimization for Recommendation Systems

T0 review · 3 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read By logging training data per user request instead of per impression, the paper claims recommender systems can cut user-side computation from one pass per impression to one pass per request.

desk verdict Credible industrial contribution: request-level training data is a genuine efficiency win, but the distribution-parity evidence is thinner than the headline claims. read the letter →

arxiv 2508.05640 v3 pith:GMGHCZ64 submitted 2025-07-24 cs.IR cs.AI

Liang Guo , Wei Li , Lucy Liao , Huihui Cheng , Rui Zhang , Yu Shi , Yueming Wang , Yanzun Huang
show 20 more authors
Keke Zhai Pengchao Wang Timothy Shi Xuan Cao Shengzhi Wang Renqin Cai Zhaojie Gong Omkar Vichare Rui Jian Leon Gao Shiyan Deng Xingyu Liu Xiong Zhang Fu Li Wenlei Xie Bin Wen Rui Li Lu Fang Xing Liu Jiaqi Zhai
This is my paper · ORCID
classification cs.IRcs.AI
keywords request-onlyoptimizationrequest-leveltrainingdatafeaturededuplicationrecommendersystemsDLRMHSTUUserArchthroughput
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

The paper tries to establish that deep learning recommendation models train better and cheaper when the unit of training data is a user request rather than a single impression. In current pipelines, one request that produces four to seven items yields four to seven nearly identical training samples, each carrying a full copy of the user's features. Collapsing these into one request-level sample removes that duplication at the source, increases usable training volume by 43% to 150% under fixed storage, and raises training throughput by 32% to 570% across production models. The authors further argue that the saved compute is what makes heavy sequential user-history architectures such as HSTU and UserArch practical to train. If right, the paper shifts the default unit of DLRM data engineering from impression-level to request-level logging, storage, and model design.

What carries the argument

The load-bearing object is the request-level training sample produced by the request-level joiner. It separates request-only (RO) features, one copy of user-side dense, id-list, and sequence features, from non-request-only (NRO) features, an array of item-side features with an impressions-per-sample count tensor. This schema does three jobs at once: it eliminates feature duplication at logging time, it lets tensor preprocessing and embedding lookup run the RO part once per request instead of once per impression, and it gives model architects a place to put expensive user-side computation, such as UserArch's Linear Compress Embedding and HSTU self-attention over the user sequence, whose cost is amortized across all impressions in the request. The reduction from $B_{NRO}$ to $B_{RO}$ examples is the mechanism that carries both the storage and the throughput claims.

What would settle it

Compare two models trained on the same production stream, one on impression-level data and one on ROO data with a shortened join window, and measure per-item-type label recall and offline normalized entropy for item categories whose feedback typically arrives after the window closes; if ROO systematically degrades recall for those categories or its normalized-entropy advantage reverses, the drop-in equivalence claim fails. The paper's Table 3 reports only conversion and video-view-duration mismatches, so a mismatch measurement on rarer, slower labels such as long-term engagement or purchase would be a direct test.

Watch

Extended reading notes

Core claim

On its own terms, this paper discovers that the whole DLRM pipeline can be restructured around a request-level sample schema in which user-side (RO) features appear once per request and item-side (NRO) features appear as an array, one entry per impression in that request. A request-level joiner buffers user-item events by request id until the join window closes, then emits a single compact example. Because user-side features dominate the feature space and are identical across impressions of one request, user-side training computation and all-to-all communication drop from $O(B_{NRO})$ to $O(B_{RO})$, with $B_{NRO}/B_{RO}$ typically 4 to 7; in production this appears as multi-fold throughput gains and as the ability to run generative-recommender-style sequential models at 7x the FLOPs under the same training compute. The paper also reports offline and online quality gains from ROO-based architectures, including up to 4.77% lift in a retrieval consumption metric and 0.04% to 0.25% topline lifts in late-stage ranking.

Load-bearing premise

The load-bearing premise is that joining at the request level, waiting roughly 16 minutes for a request's items to arrive, does not systematically change which labels and user signals appear in the training data; if late-arriving feedback is concentrated on particular item types, the efficiency gains could come with a quality loss that the aggregate mismatch rates of 0.01% to 1.07% would miss.

Editorial extensions

If this is right

  • Retrieval and early-stage ranking models, whose two-tower design puts most cost in the user tower, can see 220% to 570% training-throughput gains from ROO with minimal code changes.
  • Late-stage ranking models gain 32% to 100% throughput, and the freed compute can be spent on user-side compression (UserArch) and sequential user modeling (HSTU) that improve offline normalized entropy and online metrics.
  • The same storage budget holds 43% to 150% more training samples, reducing the need for downsampling in data logging.
  • ROO unifies training and inference feature preprocessing, removing client-side user-feature broadcast and server-side deduplication in serving.
  • ROO expansion adapters provide backward compatibility, letting existing impression-level models consume ROO data while a fleet migrates.

Reading between the lines

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

  • If the join-window distributional parity claim generalizes beyond the measured labels, conversion and video view duration, ROO could also change how online-training freshness is budgeted, since the request-level join window of about 16 minutes becomes a tunable freshness and latency knob rather than a fixed logging artifact.
  • The same amortization argument should transfer to any multi-candidate training or inference setting where one context is scored against many items, such as ads, search, and feed surfaces, so the 4-7 impressions-per-request ratios reported here may understate gains on surfaces with longer candidate lists.
  • A testable extension is to combine ROO with target-aware sequence selection methods: request-level amortization removes one of the main cost objections to per-candidate attention, so the comparison between amortized pairwise attention and HSTU-style self-attention becomes an empirical question the paper does not settle.
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

3 major / 5 minor

Summary. The paper proposes Request-Only Optimization (ROO), a training-data and modeling paradigm for industry-scale DLRMs. Instead of logging one training sample per impression, ROO logs one sample per user request, storing user-side (RO) features once and item-side (NRO) features as an array. The authors argue this removes feature duplication at the source, increases the number of training samples storable under fixed storage, reduces user-side compute and communication from O(B_NRO) to O(B_RO), and enables scaled-up user-side architectures such as UserArch and HSTU. They report production results across several Meta products: 43% to 150% more training samples, 32% to 570% training-throughput gains, and offline/online quality improvements for retrieval, early-stage, and late-stage ranking models. The paper includes a request-level join algorithm, data-quality mismatch tables, and implementation appendices.

Significance. If the reported results hold, ROO is a significant industrial contribution: it identifies a pervasive duplication inefficiency in impression-level training data, provides a clean data-format fix, and demonstrates large efficiency gains across multiple billion-user products. The core deduplication argument is simple and correct arithmetic: with 4-7 impressions per request, user-side computation is reduced from B_NRO to B_RO examples, and the amortization calculation for sequential encoders in Section 3.3 is explicit and parameter-free. The paper also gives a concrete pseudocode join algorithm and deployment details, which strengthens reproducibility of the method itself, even though the production code and data are proprietary. The main risk is that the central 'drop-in quality parity' claim rests on aggregate statistics that may not capture systematic biases in the request-level join, and the quality experiments do not cleanly isolate the data-format effect from the architectural changes it enables.

major comments (3)
  1. [Section 2.1.3 and Algorithm 1] The claim that ROO data preserves the impression-level training distribution is not sufficiently supported. Section 2.1.3 and Table 3 report only aggregate label mismatch rates (0.01% to 1.07%) for conversion and video-view-duration labels. These aggregates can hide systematic bias for rare item types, long sessions, or delayed feedback. Algorithm 1 closes the join window on fixed time or engagement thresholds, so a long user session is split into multiple ROO records and any feedback arriving after close is either lost or deferred to a different sample. No per-slice analysis (e.g., by session length, item type, or feedback latency) is provided. This matters because the paper's efficiency gains are only valuable if the resulting training data does not change what the model learns; the current evidence is too coarse to establish that.
  2. [Section 2.1.2 vs. Appendix B] There is a direct contradiction about data-landing latency. Section 2.1.2 states that request-level joining 'could potentially increase activity-to-training latency' and that ROO samples' data landing latency is about half the fixed-time join window. Appendix B, however, asserts that 'ROO data does not require additional data infra feature, nor does it add any data landing latency increase.' These statements cannot both be true without clarification. If the fixed-time join window can be as long as tens of minutes, then ROO adds latency relative to impression-level logging, which is material for freshness-sensitive models. The paper should reconcile these statements and report the actual fixed-time window values and the resulting activity-to-training latency increase, not just an average 16-minute gap between the first and last event in a request.
  3. [Section 4.3 and Tables 6-7] The quality experiments do not isolate the effect of the ROO data format from the effect of the new architectures or the larger training sample volume. Table 6 compares HSTU (Impression) at 6.8x FLOPs per example with HSTU (ROO) at 0.99x, but there is no matched-compute comparison of the same architecture trained on impression-level data versus ROO data. Table 7 reports lifts from UserArch and HSTU on top of production models, but these architectures are enabled by ROO's efficiency, so the improvements could come from the architecture, the data format, the increased sample volume (Table 4), or any combination. To support the paper's stronger claim that ROO data is quality-neutral or quality-improving relative to impression data, the authors should include a same-architecture, matched-compute comparison (e.g., impression-level data expanded from ROO samples, or ROO data expanded with the Appendix C adapter) versus native impression-level data.
minor comments (5)
  1. [Algorithm 1] The procedure name 'ReqestLevelJoin' is misspelled; it should be 'RequestLevelJoin'.
  2. [Appendix B] The custom tensor name is spelled 'InverseKeyedJaggeredTensor' in Appendix B but 'InverseKeyedJaggedTensors' in Section 5; please unify the spelling.
  3. [Section 4.3.2] The sentence 'where 0.24% and 0.03% are significant, respectively' is ambiguous because the preceding sentence lists three metrics (consumption, topline) and two percentages; please state which significance threshold applies to which metric.
  4. [Section 2.1.2] The phrase 'data landing latency is about half the fixed-time join window' needs a definition of the fixed-time join window and typical values; otherwise the reader cannot assess the freshness impact.
  5. [ACM Reference Format] The reference-format block still contains the placeholder dates '2018' and 'Conference acronym ’XX'; this should be updated for the actual submission venue.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: ROO's storage and throughput gains follow directly from request-level deduplication arithmetic, and the HSTU self-citation is not load-bearing for those central claims.

full rationale

The central efficiency claims are derived from the ROO data format itself rather than from fitted parameters or from results that presuppose the conclusion. The reduction in computational workload from B_NRO to B_RO examples is a direct arithmetic consequence of grouping impressions by request and computing user-side features once per request; it is not a prediction that is later validated by the same formula. Similarly, the reported 43-150% training sample volume increases under fixed storage follow from removing duplicated user features at logging time, and the throughput gains in Table 5 are measured system benchmarks, not quantities forced by a fitted parameter. The request-level join quality claim (Section 2.1.3) is supported by aggregate label mismatch rates in Table 3; this is an empirical parity check rather than a circular definition, though it is a distributional assumption worth scrutinizing separately. The main self-citation concern is HSTU [42], which is prior work by overlapping authors and is used as the scaled sequential architecture for quality experiments. Those quality lifts are not independent of the authors' prior architecture work, but the paper also reports its own offline and online evaluations of HSTU and UserArch (Tables 6 and 7), and the ROO storage and throughput claims do not require HSTU's quality to hold. No equation was found to be equal to its own input by construction, no fitted parameter was renamed as a prediction, and no uniqueness theorem was imported by self-citation to forbid alternatives. The latency-related statements in Section 2.1.2 and Appendix B are in some tension, but that is a consistency concern, not circularity.

Assumptions & free parameters 0 free parameters · 5 assumptions · 0 invented entities

The ROO paradigm itself is the proposed artifact; it introduces no fitted constants or invented entities. Its claims rest on standard complexity formulas and on domain assumptions about request-level feature identity, join-window label parity, and attribution of production metrics.

assumptions (5)
  • standard math Self-attention complexity is O(n^2 d + n d^2) for a sequence of length n and hidden dimension d.
    Used in Section 3.3 to compute the computational saving from deduplicating user sequence processing across the m impressions in a request.
  • domain assumption All impressions in a request share identical user-side features, so the user representation can be computed once and reused for every item.
    This is the foundation of ROO data (Section 2.1) and user-tower deduplication (Section 3.1). It is definitionally true for a logged request, but the paper assumes feature values do not depend on the item being scored.
  • domain assumption Closing the request join window after an average of 16 minutes does not materially shift the training label distribution.
    Section 2.1.2-2.1.3 asserts latency is acceptable and Table 3 reports small label mismatch rates (0.01%-1.07%). This is load-bearing for the claim that ROO data has parity with impression-level data.
  • domain assumption The reported training throughput improvements are caused by the ROO changes, not by uncontrolled infrastructure or architecture variation.
    Section 4.1 states that only ROO data, preprocessing, and pipeline were varied, but the production benchmark cannot be independently audited.
  • domain assumption Reinvesting the amortized compute into larger user-side architectures (HSTU, UserArch) yields model quality improvements.
    Sections 4.3-4.4 treat offline NE and online metric lifts as evidence for this; it is empirically tested but not guaranteed by the deduplication logic alone.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Request-Only Optimization for Recommendation Systems." pith.science (2026). https://pith.science/paper/GMGHCZ64

@misc{pith2026250805640,
  author       = {Pith},
  title        = {Pith review of: Request-Only Optimization for Recommendation Systems},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GMGHCZ64}},
  note         = {Machine review of arXiv:2508.05640}
}
read the original abstract

Deep Learning Recommendation Models (DLRMs) represent one of the largest machine learning applications on the planet. Industry-scale DLRMs are trained with petabytes of recommendation data to serve billions of users every day. To utilize the rich user signals in the long user history, DLRMs have been scaled up to unprecedented complexity, up to trillions of floating-point operations (TFLOPs) per example. This scale, coupled with the huge amount of training data, necessitates new storage and training algorithms to efficiently improve the quality of these complex recommendation systems. In this paper, we present a Request-Only Optimizations (ROO) training and modeling paradigm. ROO simultaneously improves the storage and training efficiency as well as the model quality of recommendation systems. We holistically approach this challenge through co-designing data (i.e., request-only data), infrastructure (i.e., request-only based data processing pipeline), and model architecture (i.e., request-only neural architectures). Our ROO training and modeling paradigm treats a user request as a unit of the training data. Compared with the established practice of treating a user impression as a unit, our new design achieves native feature deduplication in data logging, consequently saving data storage. Second, by de-duplicating computations and communications across multiple impressions in a request, this new paradigm enables highly scaled-up neural network architectures to better capture user interest signals, such as Generative Recommenders (GRs) and other request-only friendly architectures.

Figures

Figures reproduced from arXiv: 2508.05640 by the authors.

Figure 1
Figure 1. A traditional training data pipeline incurs excessive feature duplication at the source of data generation, i.e. impression [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Number of items (i.e., number of impression-level [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Compared to impression training, ROO training [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: The two-tower like models in the recommendation system with lightweight interaction between the two towers, e.g. [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: Sequential modeling and UserArch. Training samples increase Product-A 43% Product-B 150% Product-C 100% [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 6
Figure 6. Figure 6: The late stage ranking model architecture. HSTU and user arch (highlighted) introduced request only computations [PITH_FULL_IMAGE:figures/full_fig_p012_6.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

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

  1. ROCS: Request-Oriented Compute Sharing for Efficient Large-Scale Recommendation

    cs.LG 2026-07 conditional novelty 7.0 of 10

    ROCS restructures recommendation models so user-side computation is shared across all candidate items, yielding up to 3x serving throughput at equal or better prediction quality.

  2. MixFormer: Co-Scaling Up Dense and Sequence in Industrial Recommenders

    cs.IR 2026-02 conditional novelty 6.0 of 10

    MixFormer unifies dense feature interaction and user-sequence modeling in a single Transformer-style backbone with a user-item decoupling speedup, reporting accuracy and efficiency gains over stacked and parallel reco...

Reference graph

Works this paper leans on

52 extracted references · 24 canonical work pages · cited by 2 Pith papers

  1. [1]

    Fedor Borisyuk, Lars Hertel, Ganesh Parameswaran, Gaurav Srivastava, Sudar- shan Srinivasa Ramanujam, Borja Ocejo, Peng Du, Andrei Akterskii, Neil Daftary, Shao Tang, Daqi Sun, Qiang Charles Xiao, Deepesh Nathani, Mohit Kothari, Yun Dai, and Aman Gupta. 2025. From Features to Transformers: Redefining Ranking for Scalable Impact. arXiv:2502.03417 [cs.LG] h...

    work page arXiv 2025

  2. [2]

    Yue Cao, Xiaojiang Zhou, Jiaqi Feng, Peihao Huang, Yao Xiao, Dayao Chen, and Sheng Chen. 2022. Sampling is all you need on modeling long-term user behaviors for CTR prediction. InProceedings of the 31st ACM International Conference on Information & Knowledge Management. 2974–2983

    2022

  3. [3]

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

    2023

  4. [4]

    Edgar F Codd. 1970. A relational model of data for large shared data banks. Commun. ACM13, 6 (1970), 377–387

    work page 1970

  5. [5]

    Edgar F Codd. 1972. Further normalization of the data base relational model. Data base systems6 (1972), 33–64

    work page 1972

  6. [6]

    Paul Covington, Jay Adams, and Emre Sargin. 2016. Deep Neural Networks for YouTube Recommendations. InProceedings of the 10th ACM Conference on Recommender Systems (RecSys ’16). 191–198

    work page 2016

  7. [7]

    Bailu Ding and Jiaqi Zhai. 2025. Retrieval with Learned Similarities. InPro- ceedings of the ACM on Web Conference 2025(Sydney NSW, Australia)(WWW ’25). Association for Computing Machinery, New York, NY, USA, 1626–1637. doi:10.1145/3696410.3714822

    arXiv 2025

  8. [8]

    Zhichao Feng, JunJie Xie, Kaiyuan Li, Yu Qin, Pengfei Wang, Qianzhong Li, Bin Yin, Xiang Li, Wei Lin, and Shangguang Wang. 2024. Context-based Fast Recommendation Strategy for Long User Behavior Sequence in Meituan Waimai. InCompanion Proceedings of the ACM on Web Conference 2024. 355–363

    work page 2024

Show all 52 references
  1. [9]

    Kun Gai, Xiaoqiang Zhu, Han Li, Kai Liu, and Zhe Wang. 2017. Learning piece- wise linear models from large scale data for ad click prediction.arXiv preprint arXiv:1704.05194(2017)

    2017 arXiv

  2. [10]

    Shane Culpepper

    Luke Gallagher, Ruey-Cheng Chen, Roi Blanco, and J. Shane Culpepper. 2019. Joint Optimization of Cascade Ranking Models. InProceedings of the Twelfth ACM International Conference on Web Search and Data Mining(Melbourne VIC, Australia)(WSDM ’19). Association for Computing Machi...

    2019

  3. [11]

    Adithya Gangidi, Rui Miao, Shengbao Zheng, Sai Jayesh Bondu, Guilherme Goes, Hany Morsy, Rohit Puri, Mohammad Riftadi, Ashmitha Jeevaraj Shetty, Jingyi Yang, Shuqiang Zhang, Mikel Jimenez Fernandez, Shashidhar Gandham, and Hongyi Zeng. 2024. RDMA over Ethernet for Distributed ...

    2024

  4. [12]

    Weihao Gao, Xiangjun Fan, Chong Wang, Jiankai Sun, Kai Jia, Wenzi Xiao, Ruofan Ding, Xingyan Bin, Hui Yang, and Xiaobing Liu. 2021. Learning An End-to-End Structure for Retrieval in Large-Scale Recommendations. InProceedings of the 30th ACM International Conference on Informat...

    2021

  5. [13]

    Udit Gupta, Samuel Hsia, Vikram Saraph, Xiaodong Wang, Brandon Reagen, Gu- Yeon Wei, Hsien-Hsin S Lee, David Brooks, and Carole-Jean Wu. 2020. Deeprecsys: A system for optimizing end-to-end at-scale neural recommendation inference. In2020 ACM/IEEE 47th Annual International Sym...

    2020

  6. [14]

    Udit Gupta, Carole-Jean Wu, Xiaodong Wang, Maxim Naumov, Brandon Reagen, David Brooks, Bradford Cottel, Kim Hazelwood, Mark Hempstead, Bill Jia, et al

  7. [15]

    Ruidong Han, Bin Yin, Shangyu Chen, He Jiang, Fei Jiang, Xiang Li, Chi Ma, Mincong Huang, Xiaoguang Li, Chunzhen Jing, Yueming Han, Menglei Zhou, Lei Yu, Chuan Liu, and Wei Lin. 2025. MTGR: Industrial-Scale Gen- erative Recommendation Framework in Meituan. arXiv:2505.18654 [cs...

    2025 arXiv

  8. [16]

    Xinran He, Junfeng Pan, Ou Jin, Tianbing Xu, Bo Liu, Tao Xu, Yanxin Shi, Antoine Atallah, Ralf Herbrich, Stuart Bowers, et al. 2014. Practical lessons from predicting clicks on ads at facebook. InProceedings of the eighth international workshop on data mining for online advert...

    2014

  9. [17]

    Jui-Ting Huang, Ashish Sharma, Shuying Sun, Li Xia, David Zhang, Philip Pronin, Janani Padmanabhan, Giuseppe Ottaviano, and Linjun Yang. 2020. Embedding- based Retrieval in Facebook Search.CoRRabs/2006.11632 (2020). arXiv:2006.11632 https://arxiv.org/abs/2006.11632

    2020 arXiv

  10. [18]

    Ruiwen Xu, and Lei Zhang

    Yanhua Huang, Yuqi Chen, Xiong Cao, Rui Yang, Mingliang Qi, Yinghao Zhu, Qingchang Han, Yaowei Liu, Zhaoyu Liu, Xuefeng Yao, Yuting Jia, Leilei Ma, Yinqi Zhang, Taoyu Zhu, Liujie Zhang, Lei Chen, Weihang Chen, Min Zhu, Conference acronym ’XX, June 03–05, 2018, Woodstock, NY Gu...

    2018 arXiv

  11. [19]

    Dmytro Ivchenko, Dennis Van Der Staay, Colin Taylor, Xing Liu, Will Feng, Rahul Kindi, Anirudh Sudarshan, and Shahin Sefati. 2022. Torchrec: a pytorch domain library for recommendation systems. InProceedings of the 16th ACM Conference on Recommender Systems. 482–483

    2022

  12. [20]

    Norman P Jouppi, Cliff Young, Nishant Patil, David Patterson, Gaurav Agrawal, Raminder Bajwa, Sarah Bates, Suresh Bhatia, Nan Boden, Al Borchers, et al. 2017. In-datacenter performance analysis of a tensor processing unit. InProceedings of the 44th annual international symposi...

    2017

  13. [21]

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

    2018

  14. [22]

    William Kent. 1983. A simple guide to five normal forms in relational database theory.Commun. ACM26, 2 (1983), 120–125

    1983

  15. [23]

    Kirill Khrylchenko, Artem Matveev, Sergei Makeev, and Vladimir Baikalov

  16. [24]

    Chao Li, Zhiyuan Liu, Mengmeng Wu, Yuchi Xu, Pipei Huang, Huan Zhao, Guo- liang Kang, Qiwei Chen, Wei Li, and Dik Lun Lee. 2019. Multi-Interest Network with Dynamic Routing for Recommendation at Tmall. arXiv:1904.08030 [cs.IR] https://arxiv.org/abs/1904.08030

    2019 arXiv

  17. [25]

    Muyang Li, Xiangyu Zhao, Chuan Lyu, Minghao Zhao, Runze Wu, and Ruocheng Guo. 2022. MLP4Rec: A Pure MLP Architecture for Sequential Recommendations. InProceedings of the Thirty-First International Joint Conference on Artificial Intelli- gence, IJCAI-22, Lud De Raedt (Ed.). Int...

    2022 doi

  18. [26]

    Chi Liu, Jiangxia Cao, Rui Huang, Kai Zheng, Qiang Luo, Kun Gai, and Guorui Zhou. 2024. KuaiFormer: Transformer-Based Retrieval at Kuaishou.arXiv preprint arXiv:2411.10057(2024)

    2024 arXiv

  19. [27]

    Xiao Lv, Jiangxia Cao, Shijie Guan, Xiaoyou Zhou, Zhiguang Qi, Yaqiang Zang, Ming Li, Ben Wang, Kun Gai, and Guorui Zhou. 2025. MARM: Unlocking the Future of Recommendation Systems through Memory Augmentation and Scalable Complexity. arXiv:2411.09425 [cs.IR] https://arxiv.org/...

    2025 arXiv

  20. [28]

    Dheevatsa Mudigere, Yuchen Hao, Jianyu Huang, Zhihao Jia, Andrew Tulloch, Srinivas Sridharan, Xing Liu, Mustafa Ozdal, Jade Nie, Jongsoo Park, et al. 2022. Software-hardware co-design for fast and scalable training of deep learning rec- ommendation models. InProceedings of the...

    2022

  21. [29]

    Maxim Naumov, Dheevatsa Mudigere, Hao-Jun Michael Shi, Jianyu Huang, Narayanan Sundaraman, Jongsoo Park, Xiaodong Wang, Udit Gupta, Carole- Jean Wu, Alisson G Azzolini, et al. 2019. Deep learning recommendation model for personalization and recommendation systems.arXiv preprin...

    2019 arXiv

  22. [30]

    Nikil Pancha, Andrew Zhai, Jure Leskovec, and Charles Rosenberg. 2022. Pinner- former: Sequence modeling for user representation at pinterest. InProceedings of the 28th ACM SIGKDD conference on knowledge discovery and data mining. 3702–3712

    2022

  23. [31]

    Qi Pi, Weijie Bian, Guorui Zhou, Xiaoqiang Zhu, and Kun Gai. 2019. Practice on long sequential user behavior modeling for click-through rate prediction. InProceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. 2671–2679

    2019

  24. [32]

    Qi Pi, Guorui Zhou, Yujing Zhang, Zhe Wang, Lejian Ren, Ying Fan, Xiaoqiang Zhu, and Kun Gai. 2020. Search-based user interest modeling with lifelong sequential behavior data for click-through rate prediction. InProceedings of the 29th ACM International Conference on Informati...

    2020

  25. [33]

    Tran, Jonah Samost, Maciej Kula, Ed H

    Shashank Rajput, Nikhil Mehta, Anima Singh, Raghunandan Hulikal Keshavan, Trung Vu, Lukasz Heldt, Lichan Hong, Yi Tay, Vinh Q. Tran, Jonah Samost, Maciej Kula, Ed H. Chi, and Maheswaran Sathiamoorthy. 2023. Recommender Systems with Generative Retrieval. InThirty-seventh Confer...

    2023

  26. [34]

    Zihua Si, Lin Guan, ZhongXiang Sun, Xiaoxue Zang, Jing Lu, Yiqun Hui, Xingchao Cao, Zeyu Yang, Yichen Zheng, Dewei Leng, et al. 2024. Twin v2: Scaling ultra- long user behavior sequence modeling for enhanced ctr prediction at kuaishou. InProceedings of the 33rd ACM Internation...

    2024

  27. [35]

    Ilya Tolstikhin, Neil Houlsby, Alexander Kolesnikov, Lucas Beyer, Xiaohua Zhai, Thomas Unterthiner, Jessica Yung, Andreas Steiner, Daniel Keysers, Jakob Uszko- reit, Mario Lucic, and Alexey Dosovitskiy. 2021. MLP-mixer: an all-MLP archi- tecture for vision. InProceedings of th...

    2021

  28. [36]

    A Vaswani. 2017. Attention is all you need.Advances in Neural Information Processing Systems(2017)

    2017

  29. [37]

    Ruoxi Wang, Rakesh Shivanna, Derek Cheng, Sagar Jain, Dong Lin, Lichan Hong, and Ed Chi. 2021. Dcn v2: Improved deep & cross network and practical lessons for web-scale learning to rank systems. InProceedings of the web conference 2021. 1785–1797

    2021

  30. [38]

    Mark Wilkening, Udit Gupta, Samuel Hsia, Caroline Trippel, Carole-Jean Wu, David Brooks, and Gu-Yeon Wei. 2021. RecSSD: near data processing for solid state drive based recommendation inference. InProceedings of the 26th ACM International Conference on Architectural Support fo...

    2021

  31. [39]

    Jiajing Xu, Andrew Zhai, and Charles Rosenberg. 2022. Rethinking personalized ranking at Pinterest: An end-to-end approach. InProceedings of the 16th ACM Conference on Recommender Systems. 502–505

    2022

  32. [40]

    Xin You, Hailong Yang, Siqi Wang, Tao Peng, Chen Ding, Xinyuan Li, Bangduo Chen, Zhongzhi Luan, Tongxuan Liu, Yong Li, and Depei Qian. 2024. Exploiting Structured Feature and Runtime Isolation for High-Performant Recommendation Serving.IEEE Trans. Comput.73, 11 (2024), 2474–24...

    2024 doi

  33. [41]

    Jiaqi Zhai, Zhaojie Gong, Yueming Wang, Xiao Sun, Zheng Yan, Fu Li, and Xing Liu. 2023. Revisiting Neural Retrieval on Accelerators. InProceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining. 5520–5531

    2023

  34. [42]

    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

  35. [43]

    Buyun Zhang, Liang Luo, Xi Liu, Jay Li, Zeliang Chen, Weilin Zhang, Xiaohan Wei, Yuchen Hao, Michael Tsang, Wenjun Wang, et al. 2022. DHEN: A deep and hierarchical ensemble network for large-scale click-through rate prediction. arXiv preprint arXiv:2203.11014(2022)

    2022 arXiv

  36. [44]

    Luankang Zhang, Kenan Song, Yi Quan Lee, Wei Guo, Hao Wang, Yawen Li, Huifeng Guo, Yong Liu, Defu Lian, and Enhong Chen. 2025. Killing Two Birds with One Stone: Unifying Retrieval and Ranking with a Single Generative Rec- ommendation Model. arXiv:2504.16454 [cs.IR] https://arx...

    2025 arXiv

  37. [45]

    Mark Zhao, Niket Agarwal, Aarti Basant, Buğra Gedik, Satadru Pan, Mustafa Ozdal, Rakesh Komuravelli, Jerry Pan, Tianshu Bao, Haowei Lu, et al . 2022. Understanding data storage and ingestion for large-scale deep recommendation model training: Industrial product. InProceedings ...

    2022

  38. [46]

    Mark Zhao, Dhruv Choudhary, Devashish Tyagi, Ajay Somani, Max Kaplan, Sung-Han Lin, Sarunya Pumma, Jongsoo Park, Aarti Basant, Niket Agarwal, et al

  39. [47]

    Guorui Zhou, Na Mou, Ying Fan, Qi Pi, Weijie Bian, Chang Zhou, Xiaoqiang Zhu, and Kun Gai. 2019. Deep interest evolution network for click-through rate prediction. InProceedings of the AAAI conference on artificial intelligence, Vol. 33. 5941–5948

    2019

  40. [48]

    Lixi Zhou, Jiaqing Chen, Amitabh Das, Hong Min, Lei Yu, Ming Zhao, and Jia Zou. 2022. Serving deep learning models with deduplication from relational databases.arXiv preprint arXiv:2201.10442(2022)

    2022 arXiv

  41. [49]

    Chandler Zuo, Jonathan Castaldo, Hanqing Zhu, Haoyu Zhang, Ji Liu, Yangpeng Ou, and Xiao Kong. 2024. Inductive Modeling for Realtime Cold Start Recom- mendations. InProceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining(Barcelona, Spain)(KDD ’24)....

    2024

  42. [2020]

    In2020 IEEE International Symposium on High Performance Computer Architecture (HPCA)

    The architectural implications of facebook’s dnn-based personalized recom- mendation. In2020 IEEE International Symposium on High Performance Computer Architecture (HPCA). IEEE, 488–501

  43. [2023]

    RecD: Deduplication for end-to-end deep learning recommendation model training infrastructure.Proceedings of Machine Learning and Systems5 (2023), 754–767

    2023

  44. [2025]

    arXiv:2507.15994 [cs.IR] https://arxiv.org/abs/2507.15994

    Scaling Recommender Transformers to One Billion Parameters. arXiv:2507.15994 [cs.IR] https://arxiv.org/abs/2507.15994

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.