Pith. sign in

REVIEW 4 major objections 4 minor 10 cited by

KuaiFormer: Transformer-Based Retrieval at Kuaishou

T0 review · 4 major / 4 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read The paper claims a pure-Transformer retrieval model, KuaiFormer, lifted short-video watch time in Kuaishou's three largest scenarios after going live in May 2024.

desk verdict A credible industrial Transformer-retrieval deployment with a clear architecture, but the empirical claims are underdetermined: no held-out evaluation, no significance tests, and no control for the extra candidate supply in the online A/B. read the letter →

arxiv 2411.10057 v1 pith:6XVUN5UG submitted 2024-11-15 cs.IR cs.AIcs.LG

Chi Liu , Jiangxia Cao , Rui Huang , Kai Zheng , Qiang Luo , Kun Gai , Guorui Zhou This is my paper · ORCID
classification cs.IRcs.AIcs.LG
keywords short-videorecommendationtransformeruserinterestmodelingmulti-interestquerytokensadaptiveitemcompressionnextactionpredictionin-batchsoftmaxindustrialretrieval
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

KuaiFormer is a retrieval-stage model for short-video recommendation that replaces the usual score-estimation task (predicting click-through rate) with a transformer-driven Next Action Prediction task: given a user's recently watched videos, it predicts which video the user will engage with next. The paper argues that this shift, together with multiple learnable query tokens that extract several distinct interest vectors and an adaptive compression mechanism that coarsens older watched items, lets a pure Transformer architecture work at industrial scale for the first time in retrieval. The model has been live in the Kuaishou app since May 2024, serving over 400 million daily active users, and online A/B tests in the three largest scenarios report video watch time gains of +0.360%, +0.126%, and +0.411%. A sympathetic reader would care because the paper claims to close the gap between academic Transformer recommenders and the efficiency and scale constraints of real deployments, where sequence length and billion-item candidate pools usually force simpler dual-tower models.

What carries the argument

The load-bearing machinery is an autoregressive Transformer backbone (Llama-style, with RMSNorm, masked multi-head self-attention, and a pointwise feed-forward layer) that consumes item tokens and, appended after them, a set of learnable query tokens under the same causal mask, so each successive interest vector can attend to earlier interest vectors and to the whole item sequence. An adaptive compression stage splits the behavior history into early, middle, and latest parts, compresses groups of 64 and 16 old items via a single-layer bidirectional Transformer followed by mean pooling, and concatenates the compressed tokens with the latest uncompressed items; this turns a 256-item history into 55 tokens and is what keeps quadratic self-attention affordable. The training objective is a smoothed in-batch softmax with logQ correction and label smoothing, which stabilizes learning over a billion-scale candidate set and tolerates noisy short-video labels. Together these mechanisms carry the paper's three claims: multi-interest extraction, long-sequence efficiency, and stable billion-item training.

What would settle it

Re-run the Section 4.4 hyperparameter comparisons (sequence length, query-token count, layer depth, compression strategy) using held-out next-item prediction or online hit rate on logs from a different day; if the compressed 256-item model does not beat the uncompressed 256-item model on held-out data, the claim that compression reduces noise rather than merely improving training fit is refuted.

Watch

Extended reading notes

Core claim

The central claim is that retrieval can be redefined as Next Action Prediction with a pure Transformer encoder instead of a score-estimation task such as click-through rate prediction. Given the user's recent watched-video sequence, KuaiFormer produces several user interest vectors from learnable query tokens; the score for a candidate video is the maximum inner product between the video embedding and those interest vectors. To make this tractable, the paper introduces adaptive item compression, which groups and mean-pools older items through a one-layer bidirectional Transformer, and a smooth in-batch softmax loss with logQ correction, which avoids computing over the full billion-item vocabulary. In offline replay of real requests, KuaiFormer reports the highest hit rate among deployed baselines (for example, HR@1000 of 19.88% versus 10.26% for the GNN pathway), and the paper states that this is the first real-time retrieval model built on a pure Transformer architecture at industrial scale.

Load-bearing premise

The paper's offline evidence for its design choices rests on 'Accuracy' measured as the model's fit to its own training data once the loss stops falling, not on held-out user behavior; if that training fit does not track retrieval quality on unseen logs, the claimed benefits of query tokens and compression are not demonstrated by the offline experiments.

Editorial extensions

If this is right

  • Retrieval no longer needs a dual-tower interaction at the top layer; a single Transformer can produce user vectors that are compared with item embeddings by inner product, so the same ANN infrastructure used for embedding retrieval remains usable.
  • Longer behavior histories become affordable: compressing 256 items into 55 tokens costs roughly 10% extra compute over a 64-item baseline, so industrial systems can extend context without paying the full quadratic attention price.
  • Multi-interest is expressed as several output vectors rather than one, which increases retrieval coverage and, according to the paper, improves the 'Novel Surprise' metric that measures discovery of new interests.
  • The Next Action Prediction objective with smoothed in-batch softmax provides a recipe for training Transformer recommenders on billion-scale item sets without full softmax over the candidate pool.
  • Deployment is feasible as an additional retrieval pathway alongside existing item2item, graph, and multi-interest pathways, with minute-level online updates and GPU brute-force TopK serving.

Reading between the lines

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

  • The causal masking between query tokens is a testable design choice: an ablation with independent, non-causal query tokens would isolate whether the reported gains come from interest disentanglement through autoregressive interaction or merely from having several output vectors.
  • The compression schedule (64/16 grouping with the latest 48 items uncompressed) is reported for one configuration; whether it generalizes to other session lengths or video categories is open, since the memory-decay assumption it rests on is not directly measured.
  • One could make the number of query tokens and the compression window adaptive to each user's recent activity, since the paper fixes both as global hyperparameters; diversity of recent views would be a natural signal for choosing how many interest vectors to output.
  • The same Next Action Prediction recipe with multiple query tokens and compressed histories should transfer to other feed-style products such as e-commerce or long-video recommendation, where the bottlenecks are the same: billion-scale item sets and multiple simultaneous interests.
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 / 4 minor

Summary. The paper proposes KuaiFormer, a Transformer-based retrieval model for Kuaishou's short-video recommendation system. The architecture replaces the conventional dual-tower EBR score-estimation setup with a Next Action Prediction paradigm: an MLP/embedding layer produces token representations of watched videos, a Llama-style causal Transformer processes a compressed sequence, and multiple learnable query tokens produce several interest representations. The training objective combines in-batch softmax, logQ sampling correction, and label smoothing. The authors claim deployment since May 2024, serving over 400 million daily active users, with offline hit-rate improvements over six baselines and online A/B gains of +0.360%, +0.126%, and +0.411% in video watch time in the three largest scenarios. The paper also presents hyperparameter studies for sequence length, number of query tokens, number of layers, and item compression.

Significance. If the central claims were fully evidenced, this would be a valuable industrial case study: it demonstrates a pure Transformer retrieval model at billion-scale candidate-pool size, with a plausible treatment of multi-interest extraction and long-sequence efficiency. The detailed architecture description and deployment pipeline are useful to practitioners. However, the evaluation evidence as presented is not sufficient to establish the architectural attribution: the offline replay table lacks interval estimates, the hyperparameter analysis relies on training-set accuracy, and the online A/B section does not control for candidate-supply effects. The paper's stated contribution ('first real-time retrieval model by pure Transformer in industrial-scale RecSys') is a strong claim that needs stronger evidence.

major comments (4)
  1. [§4.1.2, §4.4, Table 3] The Offline Accuracy metric is defined as the accuracy when training loss stabilizes, which measures fit to the training behavior stream rather than held-out generalization. RQ3 and the claims about sequence length, query tokens, layer count, and the assertion that compression 'reduces noise' are based entirely on this metric (Figure 3 and Table 3). Training accuracy can rise with model capacity or sequence length without implying better retrieval on unseen requests, especially under in-batch softmax training. I request an evaluation on a temporally disjoint held-out log (e.g., next-item prediction with top-K recall) before these architectural conclusions are treated as established.
  2. [Table 1, §4.2] Table 1 reports only point estimates of HR@50 through HR@1000 for KuaiFormer and the six baselines, with no confidence intervals, number of requests, or significance tests. The headline comparisons (e.g., 26.30% relative improvement over the runner-up at HR@1000) could be within noise. The paper should report interval estimates across users or days and significance tests for the difference with the runner-up.
  3. [§4.3, Table 2] The online A/B section gives only relative percentage lifts and no p-values, confidence intervals, sample sizes, or variance measures, despite the statement that a ~0.1% video-watch-time change is 'statistically significant'. Moreover, because §3 states that KuaiFormer was added as a new retrieval pathway and candidates from all pathways are aggregated and deduplicated before ranking, the reported gains may reflect the added candidate supply rather than the Transformer architecture itself. The treatment should be compared against a control with an equal-sized additional baseline pathway or with the total candidate count and downstream ranking held fixed.
  4. [Eq. (10), Eq. (11)] Equation (10) defines Score_{x_{n+1}} as argmax({x^T u_1, ..., x^T u_k}); argmax returns an index rather than a score value and is non-differentiable, which is inconsistent with using Score in the in-batch softmax loss in Eq. (11). If the intended operation is max or a differentiable approximation, this should be stated explicitly; as written, the training objective is not well defined.
minor comments (4)
  1. [Throughout] There are several typos and formatting issues, including 'recomenndation' in §3, 'thelargest' in §4.1.1, and broken equation/table rendering in Eq. (10) and Table 3. A careful copyedit is needed.
  2. [§2.3, Eq. (7)] The compression split is not stated precisely: the text says the sequence is divided into earlier, middle, and latest parts, and Eq. (7) uses group sizes 64 and 16, but the total of the displayed groups (128 + 80 = 208) and the number of uncompressed items (48) should be tied explicitly to the final compressed length of 55 in Figure 1.
  3. [§2.2.1, Eq. (4)] The bucketing formula for duration uses a 1000-bucket uniform strategy with maximum 300s, but the paper does not state how other continuous attributes are bucketed or whether each attribute has its own bucket count. Clarifying this would improve reproducibility.
  4. [§2.5, Eq. (12)] The label smoothing loss in Eq. (12) introduces alpha but the paper does not state its chosen value or sensitivity to it; since the abstract and method sections emphasize stable training, a short ablation or at least a stated default would be helpful.

Circularity Check

1 steps flagged · score 2.0 of 10

No load-bearing circularity: the online A/B result is an independent behavioral measurement; the only circular-adjacent element is the in-sample 'Offline Accuracy' metric used to justify offline component choices.

  1. fitted input called prediction [Section 4.1.2 (Evaluation Protocol) and Section 4.4.4 (Item compression strategy impact)]
    "For evaluating the performance of our KuaiFormer variants, since the models are trained in an in-batch setting and use online learning to continuously incorporate the latest logs, the accuracy when the training loss stabilizes with the same batch size can assess the model's capability. This offline training accuracy reflects the model's ability to fit user behavior. ... Our findings indicate that the compressed sequence outperforms the model with a sequence length of 64 and, notably, achieves slightly better accuracy than the model with the full uncompressed sequence of 256 items."

    By the paper's own definition, this 'Offline Accuracy' is the converged value of the training objective, i.e., an in-sample fit measure on the training distribution. Section 4.4 tunes sequence length, query-token count, layer depth, and the compression strategy on this same metric, and then reports the resulting improvements as evidence that the components 'provide more context' or 'reduce noise.' Because the evaluation quantity is, by construction, the same objective minimized during training, the observed gains are an in-sample consequence of the fitted configuration rather than an independent prediction. This issue is localized: it does not enter the online watch-time A/B results, which are measured externally.

full rationale

The paper's central claim — that KuaiFormer improved video watch time by +0.360%/+0.126%/+0.411% in online A/B tests — is not derived from, or equivalent to, any fitted input. The training loss, architecture, and deployment are specified independently, and the online metrics are external behavioral measurements. The replay-based 'online Hit Rate' in Table 1 is an offline recall-style measure, not a construction-level circularity. Baselines such as GPRP and DimeRec include overlapping authors, but they are used as empirical comparison points, not as cited justification for KuaiFormer's architecture. The only close call is the Offline Accuracy definition, which measures training-fit convergence rather than held-out generalization; using it to validate component choices is a mild self-confirmation. Because the principal online result remains independent, the overall circularity score is low.

Assumptions & free parameters 7 free parameters · 6 assumptions · 0 invented entities

The paper introduces no new physical entities. Its central claim rests mainly on domain assumptions about user behavior (vague memory of old items, tolerance for many positives, multi-interest structure) and on hand-chosen hyperparameters, several of which are not reported. The compression sizes and smoothing coefficient are particularly load-bearing because they are selected using training accuracy rather than held-out evaluation.

free parameters (7)
  • Label smoothing coefficient alpha = not reported
    Controls the smoothing term in Eq. (12); chosen by hand, no value given.
  • Number of query tokens k = 6 (optimal in Section 4.4.2)
    Selected by accuracy/compute trade-off on internal data.
  • Number of Transformer layers L = 4 or 5 (Section 4.4.3)
    Selected by accuracy/compute trade-off; gains level off after layer 4.
  • Input sequence length n = 256 (Section 4.4.1)
    Longer sequences improve accuracy with diminishing returns; 256 chosen as a practical point.
  • Compression group sizes = 64, 64, 16, 16, 16, 16, 16, with latest 48 items uncompressed
    Hand-set in Section 4.4.4; no systematic search reported.
  • Embedding dimension d and number of attention heads M = not reported
    Core architectural hyperparameters are not disclosed.
  • Duration bucket count and maximum = 1000 buckets, max 300 seconds
    Hand-chosen bucketing strategy in Eq. (4); no sensitivity analysis.
assumptions (6)
  • standard math Self-attention complexity is O(n^2 d), making long sequences expensive.
    Used in Section 2.3 to motivate item compression; cites the standard complexity result.
  • domain assumption Users remember earlier videos more vaguely than recent videos, so coarse grouping of older items is acceptable.
    Stated as an intuitive assumption in Section 2.3, directly motivating the compression mechanism.
  • domain assumption Short-video users have high tolerance, so sampled negatives in in-batch training are not reliably negative.
    Invoked in Section 2.5 to justify label smoothing; if false, smoothing would be unnecessary or harmful.
  • domain assumption Multiple semantically different short videos can be equally valid positive next items for a user.
    Used in Section 1 and Section 2.4 to justify multi-interest query tokens.
  • domain assumption In-batch softmax with logQ correction is a faithful approximation to full softmax over a billion-scale item set.
    Adopted in Section 2.5 from prior work [35]; the approximation quality is not independently verified here.
  • domain assumption The online A/B test is a clean comparison with only KuaiFormer varying.
    Required for interpreting Table 2 as causal evidence; no details of experiment isolation are given.

how reviews work

0 comments
Cite this review

Pith. "Pith review of KuaiFormer: Transformer-Based Retrieval at Kuaishou." pith.science (2026). https://pith.science/paper/6XVUN5UG

@misc{pith2026241110057,
  author       = {Pith},
  title        = {Pith review of: KuaiFormer: Transformer-Based Retrieval at Kuaishou},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6XVUN5UG}},
  note         = {Machine review of arXiv:2411.10057}
}
read the original abstract

In large-scale content recommendation systems, retrieval serves as the initial stage in the pipeline, responsible for selecting thousands of candidate items from billions of options to pass on to ranking modules. Traditionally, the dominant retrieval method has been Embedding-Based Retrieval (EBR) using a Deep Neural Network (DNN) dual-tower structure. However, applying transformer in retrieval tasks has been the focus of recent research, though real-world industrial deployment still presents significant challenges. In this paper, we introduce KuaiFormer, a novel transformer-based retrieval framework deployed in a large-scale content recommendation system. KuaiFormer fundamentally redefines the retrieval process by shifting from conventional score estimation tasks (such as click-through rate estimate) to a transformer-driven Next Action Prediction paradigm. This shift enables more effective real-time interest acquisition and multi-interest extraction, significantly enhancing retrieval performance. KuaiFormer has been successfully integrated into Kuaishou App's short-video recommendation system since May 2024, serving over 400 million daily active users and resulting in a marked increase in average daily usage time of Kuaishou users. We provide insights into both the technical and business aspects of deploying transformer in large-scale recommendation systems, addressing practical challenges encountered during industrial implementation. Our findings offer valuable guidance for engineers and researchers aiming to leverage transformer models to optimize large-scale content recommendation systems.

Figures

Figures reproduced from arXiv: 2411.10057 by the authors.

Figure 1
Figure 1. KuaiFormer architecture under the length setting 256, and 4 query tokens, where the [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Deployment Architecture our KuaiFormer learning objective. L(𝑥𝑛+1) = − log  𝑒 Score𝑥𝑛+1 𝑒 Score𝑥𝑛+1 + ÍB 𝑥¯𝑛+1 𝑒 Score𝑥¯𝑛+1  (11) where the B denotes the in-batch negative short-video set, the 𝑥𝑛+1 denotes the positive item and the 𝑥¯𝑛+1 indicates an arbitrary negative item. However, the in-batch softmax will inevitably introduce the sampling bias to away from uniform item sampling: the top popular items have more… view at source ↗
Figure 3
Figure 3. Impact of Sequence Length, Query token number and Layer number on Accuracy [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 10 Pith papers

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

  1. WatchLens: A Configurable Platform for Online Video Recommendation Experiments

    cs.IR 2026-08 conditional novelty 6.0 of 10

    WatchLens is an open-source platform that attaches exposure context to every playback event, enabling controlled online experiments on how recommendation policies shape viewing behavior.

  2. Embedding Items at Scale: Comparing GNN-Based and ID-Based Item Embeddings in the Yandex Ecosystem

    cs.IR 2026-07 conditional novelty 6.0 of 10

    Pretrained GNN item embeddings outperform end-to-end ID embeddings on a small dataset, but not in two large-scale Yandex production recommender systems.

  3. Real-Time Hard Negative Sampling via LLM-based Clustering for Large-Scale Two-Tower Retrieval

    cs.IR 2026-07 unverdicted novelty 6.0 of 10

    Cluster-based real-time out-of-batch negatives drawn from LLM media embeddings outperform industry-standard negative sampling for two-tower retrieval and cut popularity bias.

  4. 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.

  5. Unifying Generative Recall and Multi-Objective Ranking in a Single Decoder-Only Sequence

    cs.IR 2026-07 conditional novelty 5.5 of 10

    A single decoder-only sequence with dual-query prefix-causal attention and ranking-side LoRA unifies generative SID recall and multi-objective ranking, with offline and online gains at Kuaishou.

  6. MISS: Multi-Modal Tree Indexing and Searching with Lifelong Sequential Behavior for Retrieval Recommendation

    cs.IR 2025-08 conditional novelty 5.0 of 10

    MISS builds a k-means index tree on interaction-supervised multi-modal embeddings and adds two behavior search units (Co-GSU, MM-GSU) plus ESU/MMoE, reporting ~30-47% relative recall gains over TDM+MMoE on Kuaishou da...

  7. Correcting the LogQ Correction: Revisiting Sampled Softmax for Large-Scale Retrieval

    cs.IR 2025-07 conditional novelty 5.0 of 10

    The paper derives and tests a refined logQ-correction loss that drops the positive item from the sampled denominator and scales each positive example by one minus its estimated model probability.

  8. Synergizing Implicit and Explicit User Interests: A Multi-Embedding Retrieval Framework at Pinterest

    cs.IR 2025-06 conditional novelty 5.0 of 10

    A production recommender framework combines a differentiable clustering module for implicit interests and conditional retrieval for explicit followed topics, deployed at Pinterest home feed.

  9. Next-User Retrieval: Enhancing Cold-Start Recommendations via Generative Next-User Modeling

    cs.IR 2025-06 conditional novelty 5.0 of 10

    A transformer trained to generate the next user for a cold-start item, matched via HNSW, improves Douyin's cold-start metrics in online A/B tests.

  10. 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

41 extracted references · 25 canonical work pages · cited by 10 Pith papers

  1. [1]

    Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, Floren- cia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, et al. 2023. Gpt-4 technical report

    2023

  2. [2]

    Dosovitskiy Alexey. 2020. An image is worth 16x16 words: Transformers for image recognition at scale. In arXiv

    work page 2020

  3. [3]

    Tom B Brown. 2020. Language models are few-shot learners. In arXiv

    work page 2020

  4. [4]

    Yukuo Cen, Jianwei Zhang, Xu Zou, Chang Zhou, Hongxia Yang, and Jie Tang

  5. [5]

    Jianxin Chang, Chenbin Zhang, Yiqun Hui, Dewei Leng, Yanan Niu, Yang Song, and Kun Gai. 2023. Pepnet: Parameter and embedding personalized network for infusing with personalized prior information. In ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD)

    2023

  6. [6]

    Paul Covington, Jay Adams, and Emre Sargin. 2016. Deep neural networks for youtube recommendations. InACM Conference on Recommender Systems (RecSys)

    2016

  7. [7]

    Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding. In NAACL, Jill Burstein, Christy Doran, and Thamar Solorio (Eds.)

    work page 2019

  8. [8]

    Ruiqi Guo, Philip Sun, Erik Lindgren, Quan Geng, David Simcha, Felix Chern, and Sanjiv Kumar. 2020. Accelerating large-scale inference with anisotropic vector quantization. In International Conference on Machine Learning. PMLR, 3887–3896

    2020

Show all 41 references
  1. [9]

    Xiangnan He, Kuan Deng, Xiang Wang, Yan Li, Yongdong Zhang, and Meng Wang. 2020. Lightgcn: Simplifying and powering graph convolution network for recommendation. In International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR)

    2020

  2. [10]

    Jonathan L Herlocker, Joseph A Konstan, and John Riedl. 2000. Explaining col- laborative filtering recommendations. In ACM conference on Computer supported cooperative work

    2000

  3. [11]

    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. In ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD)

    2020

  4. [12]

    Po-Sen Huang, Xiaodong He, Jianfeng Gao, Li Deng, Alex Acero, and Larry Heck. 2013. Learning deep structured semantic models for web search using clickthrough data. InACM International Conference on Information and Knowledge Management (CIKM)

    2013

  5. [13]

    Jeff Johnson, Matthijs Douze, and Hervé Jégou. 2019. Billion-scale similarity search with GPUs. In IEEE Transactions on Big Data

    2019

  6. [14]

    Wang-Cheng Kang and Julian McAuley. 2018. Self-attentive sequential recom- mendation. In IEEE international conference on data mining (ICDM)

    2018

  7. [15]

    Feyza Duman Keles, Pruthuvi Mahesakya Wijewardena, and Chinmay Hegde

  8. [16]

    Hyeyoung Ko, Suyeon Lee, Yoonseo Park, and Anna Choi. 2022. A survey of recommendation systems: recommendation models, techniques, and application fields. Electronics (2022)

    2022

  9. [17]

    Yehuda Koren, Steffen Rendle, and Robert Bell. 2021. Advances in collaborative filtering. Recommender systems handbook (2021)

    2021

  10. [18]

    Chao Li, Zhiyuan Liu, Mengmeng Wu, Yuchi Xu, Huan Zhao, Pipei Huang, Guoliang Kang, Qiwei Chen, Wei Li, and Dik Lun Lee. 2019. Multi-interest net- work with dynamic routing for recommendation at Tmall. In ACM International Conference on Information and Knowledge Management (CIKM)

    2019

  11. [19]

    Junnan Li, Dongxu Li, Caiming Xiong, and Steven Hoi. 2022. Blip: Bootstrapping language-image pre-training for unified vision-language understanding and generation. In International conference on machine learning

    2022

  12. [20]

    Sen Li, Fuyu Lv, Taiwei Jin, Guli Lin, Keping Yang, Xiaoyi Zeng, Xiao-Ming Wu, and Qianli Ma. 2021. Embedding-based product retrieval in taobao search. In ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD)

    2021

  13. [21]

    Wuchao Li, Rui Huang, Haijun Zhao, Chi Liu, Kai Zheng, Qi Liu, Na Mou, Guorui Zhou, Defu Lian, Yang Song, et al. 2024. DimeRec: A Unified Framework for En- hanced Sequential Recommendation via Generative Diffusion Models. In arXiv

    2024

  14. [22]

    Qiang Liu, Shu Wu, Diyi Wang, Zhaokang Li, and Liang Wang. 2016. Context- aware sequential recommendation. InIEEE international conference on data mining (ICDM)

    2016

  15. [23]

    Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, and Baining Guo. 2021. Swin transformer: Hierarchical vision transformer us- ing shifted windows. In Proceedings of the IEEE/CVF international conference on computer vision. 10012–10022

    2021

  16. [24]

    Rafael Müller, Simon Kornblith, and Geoffrey E. Hinton. 2019. When Does Label Smoothing Help?. In arXiv

    2019

  17. [25]

    Nikil Pancha, Andrew Zhai, Jure Leskovec, and Charles Rosenberg. 2022. Pin- nerFormer: Sequence Modeling for User Representation at Pinterest. In ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD)

    2022

  18. [26]

    Qi Pi, Weijie Bian, Guorui Zhou, Xiaoqiang Zhu, and Kun Gai. 2019. Practice on long sequential user behavior modeling for click-through rate prediction. In ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD)

    2019

  19. [27]

    Hui Shi, Yupeng Gu, Yitong Zhou, Bo Zhao, Sicun Gao, and Jishen Zhao. 2023. Everyone’s preference changes differently: A weighted multi-interest model for retrieval. In International Conference on Machine Learning . 31228–31242

    2023

  20. [28]

    Elena Smirnova and Flavian Vasile. 2017. Contextual sequence modeling for rec- ommendation with recurrent neural networks. InProceedings of the 2nd workshop on deep learning for recommender systems

    2017

  21. [29]

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

  22. [30]

    Chaofan Tao, Qian Liu, Longxu Dou, Niklas Muennighoff, Zhongwei Wan, Ping Luo, Min Lin, and Ngai Wong. 2024. Scaling laws with vocabulary: Larger models deserve larger vocabularies. In arXiv

    2024

  23. [31]

    Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timothée Lacroix, Baptiste Rozière, Naman Goyal, Eric Hambro, Faisal Azhar, et al. 2023. Llama: Open and efficient foundation language models. In arXiv

    2023

  24. [32]

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

    2017

  25. [33]

    Shoujin Wang, Liang Hu, Yan Wang, Longbing Cao, Quan Z Sheng, and Mehmet Orgun. 2019. Sequential recommender systems: challenges, progress and prospects. arXiv (2019)

    2019

  26. [34]

    Xiaoyong Yang, Yadong Zhu, Yi Zhang, Xiaobo Wang, and Quan Yuan. 2020. Large scale product graph construction for recommendation in e-commerce. In arXiv

    2020

  27. [35]

    Xinyang Yi, Ji Yang, Lichan Hong, Derek Zhiyuan Cheng, Lukasz Heldt, Adi- tee Ajit Kumthekar, Zhe Zhao, Li Wei, and Ed Chi. 2019. Sampling-Bias-Corrected Neural Modeling for Large Corpus Item Recommendations. In ACM Conference on Recommender Systems (RecSys)

    2019

  28. [36]

    Kai Zheng, Haijun Zhao, Rui Huang, Beichuan Zhang, Na Mou, Yanan Niu, Yang Song, Hongning Wang, and Kun Gai. 2024. Full Stage Learning to Rank: A Unified Framework for Multi-Stage Systems. In Proceedings of the ACM on Web Conference

    2024

  29. [37]

    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. In Proceedings of the AAAI conference on artificial intelligence

    2019

  30. [38]

    Guorui Zhou, Xiaoqiang Zhu, Chenru Song, Ying Fan, Han Zhu, Xiao Ma, Yanghui Yan, Junqi Jin, Han Li, and Kun Gai. 2018. Deep interest network for click-through rate prediction. In ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD)

    2018

  31. [2019]

    In ACM International Conference on Information and Knowledge Management (CIKM)

    BERT4Rec: Sequential recommendation with bidirectional encoder repre- sentations from transformer. In ACM International Conference on Information and Knowledge Management (CIKM)

  32. [2020]

    In ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD)

    Controllable multi-interest framework for recommendation. In ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD) . 2942–2951

  33. [2022]

    In arXiv

    On The Computational Complexity of Self-Attention. In arXiv

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.