Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

This paper argues that sparse click histories can be augmented with behavior sequences borrowed from similar high-active users, matched in a semantic space where item text and image features are aligned.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

MARS aligns text and image features of items, then augments sparse user sequences with retrieved sequences from similar heavy users, improving CTR prediction offline and online.

T0 review reviewed 2026-08-05 challenge →

load-bearing objection MARS has a sensible idea and real online evidence, but the offline evaluation as written does not rule out target-item leakage, so the headline gain is not yet trustworthy. the 4 major comments →

arxiv 2509.01184 v1 pith:YCP6BX7D submitted 2025-09-01 cs.IR

MARS: Modality-Aligned Retrieval for Sequence Augmented CTR Prediction

Yutian Xiao , Shukuan Wang , Binhao Wang , Zhao Zhang , Yanze Zhang , Shanqi Liu , Chao Feng , Xiang Li
show 1 more author
Fuzhen Zhuang
This is my paper
classification cs.IR
keywords CTR predictiondata augmentationmodality alignmentretrieval augmentationlow-active usersmultimodal user embeddingsStein kernelbehavior sequence
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The reading

This paper tries to establish that click-through rate prediction for users with sparse interaction histories can be improved by lending them the behavior sequences of similar high-active users—provided the similarity is measured in a semantic space in which item text and image signals have been aligned. The proposed framework, MARS, works in two stages: a Stein-kernel-based alignment step projects text and image features into a shared, entropy-regularized space and fuses them with item-ID embeddings, producing average-pooled user embeddings; a retrieval step then finds the most similar high-active user, filters that user's history by item-level cosine similarity, and prepends the filtered sequence to the low-active user's history. Offline, MARS reports higher AUC than heuristic and trainable augmentation baselines on MovieLens-1M, Amazon Beauty, and Amazon Toys, with the largest relative gain on the sparsest dataset. Online, deployed on a short-video recommendation platform, it reports consistent lifts in usage-time metrics and markedly larger lifts for the low-activity segment. If correct, this means that content semantics, not just collaborative signals, can safely transfer interest across users and that sparse users can be modeled better without training extra augmentation parameters.

Core claim

The central discovery is that sequences from high-active users can be transferred to low-active users without any additional trainable augmentation parameters, as long as retrieval is performed in a semantically aligned multimodal space. MARS first aligns text and image item features using a Stein-kernel objective: an RBF kernel similarity between modalities is maximized while an entropy regularizer derived from the Stein score function prevents representation collapse; cross-attention then fuses text, image, and item-ID embeddings, and each user is represented by the average of their fused item embeddings. At retrieval time, a low-active user is matched to the high-active user with the larg

What carries the argument

The carrying machinery is a two-stage retrieval pipeline governed by a single semantic space. Stage one builds that space: text and image features are projected to a common dimension, a Stein-kernel alignment loss maximizes the RBF similarity between the two modalities while an approximate-Stein-score entropy term spreads the embeddings out, and cross-attention fuses text, image, and item-ID signals into one item embedding; a user is the average of those embeddings over her history. Stage two spends the space: for each low-active user, the high-active user with the highest cosine similarity between normalized user embeddings is retrieved, each borrowed item is re-checked against the target u

Load-bearing premise

The load-bearing premise is that cosine similarity between average-pooled, CTR-trained multimodal user embeddings is a faithful measure of interest compatibility, so that a high-active user's filtered history genuinely helps a low-active user rather than injecting noise.

What would settle it

Run MARS side by side with a control that performs the same retrieval and filtering but replaces the borrowed sequence with a random sequence of the same length drawn from the same high-active user pool; if the control matches MARS's AUC, the semantic content of the borrowed items is not what drives the gain. A second, sharper check is to invert the alignment objective during pretraining (push text and image embeddings apart) and see whether retrieval-augmented AUC drops below the unaugmented base model; if it does not, alignment quality is irrelevant to the claimed effect.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • Data sparsity in CTR prediction can be attacked without training extra augmentation parameters: once the aligned embedding space is built, augmentation is a nearest-neighbor lookup plus a cosine-similarity filter.
  • Semantic consistency of borrowed items, not model complexity, is what drives gains: heuristic random or repeated padding mostly hurts, while filtered retrieval helps.
  • Both text and image signals plus explicit cross-modal alignment are jointly necessary; ablations removing any one component push AUC below the base model.
  • The best way to pack borrowed history depends on data sparsity: a single most-similar user sequence with filtering is safest on dense data, while combining several similar users helps in sparse, diverse catalogs.
  • Low-activity users are the main beneficiaries in production; the reported lifts on usage time and 7-day engagement metrics are much larger for that segment than for the average population.

Where Pith is reading between the lines

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

  • A testable implication the paper leaves open is that average pooling is not the only or best user aggregator; replacing it with attention-based pooling over fused item embeddings could sharpen retrieval and further lift CTR.
  • Because borrowed items are prepended without source markers, the downstream model must infer on its own that the prefix belongs to another user; adding a position or source feature could tell us whether the model relies on the borrowed prefix per se or just on longer sequences.
  • The framework's success suggests the aligned user embeddings could serve other cold-start tasks—candidate retrieval, follow prediction, or cross-scenario transfer—where interaction sparsity is equally the bottleneck; the paper notes the embeddings are reusable but does not quantify those uses.
  • The largest risk is silent mode collapse or modality dominance in the aligned space; an offline diagnostic that tracks the ratio of intra-modal to cross-modal nearest-neighbor distances under varying alignment weights would provide an early warning independent of AUC.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. MARS proposes a two-stage framework for CTR prediction under sparse user behavior. In stage 1, a Stein-kernel-based alignment module projects text and image features into a shared multimodal space and forms user embeddings by average pooling (Eq. 13). In stage 2, for each low-active user, the most similar high-active user is retrieved by cosine similarity (Eq. 15), their behavior sequence is filtered by item-user similarity (Eq. 16-17), and the filtered sequence is prepended to the low-active user's history (Eq. 18). The paper reports offline AUC gains over heuristic and trainable baselines on ML-1M, Amazon-Beauty, and Amazon-Toys (Table 2), ablation results (Table 3), and online A/B lifts at Kuaishou (Tables 4-5). The authors claim the method is deployed and code is provided.

Significance. If the results are valid, MARS offers a practically interesting approach to data sparsity in CTR by exploiting multimodal item semantics and cross-user sequence transfer, with the notable strength of a clear two-stage design, extensive ablation, and public code. The offline improvements over strong baselines and the positive online A/B results would support the claim of industrial applicability. However, the load-bearing offline evaluation may be compromised by label leakage due to an unspecified leave-one-out split, and the alignment loss is not fully specified. These issues must be resolved before the empirical claims can be accepted.

major comments (4)
  1. [§4.1.1 and §3.4, Eq. (18)] The leave-one-out split is described without any temporal constraint. The paper states 'adopt the leave-one-out strategy' but does not specify that retrieval for a target user excludes interactions after that user's test timestamp. Since high-active user histories are built from interaction logs (§3.5.1), a high-active user's sequence can contain the low-active user's held-out test item. Equation (18) directly prepends this sequence, so at evaluation time the true label item may appear in the augmented input. This can fully explain the offline AUC gains in Table 2, independent of any semantic transfer. Please specify the split precisely and re-run experiments with a strict 'no future interactions' retrieval constraint.
  2. [§3.2, Eq. (7)] The entropy regularizer is not defined. Equation (7) introduces ∇_e log q(e) as the 'approximate Stein score function' but does not give the estimator, the kernel, or the derivation. Consequently, the claim in the abstract of an 'unbiased semantic space' is unsubstantiated and the alignment module is not reproducible. Please provide the explicit Stein score estimator (e.g., spectral Stein gradient estimator) and the exact RBF kernel used, and justify the unbiasedness claim with a supporting derivation or reference.
  3. [Table 2] Asterisks indicate statistically significant improvements over the best trainable baseline (t-test, p<0.05), but no variance, standard deviations, or number of runs are reported. The t-test appears to be applied to point estimates only. Given the small AUC differences (e.g., ML-1M: 0.7813 vs 0.7793 for BASRec; Toys: 0.5901 vs 0.5894 for CL4SRec), reporting mean±std over multiple seeds and a paired-test setup is essential to support the significance claims.
  4. [§4.5, Tables 4-5] The online A/B test description is too thin to assess. The paper reports relative lifts but no p-values, confidence intervals, sample sizes, control/allocation details, or exact metric definitions. The claim that these are 'statistically significant improvements' is not verifiable from the provided information. Please add standard A/B reporting details, or explicitly state that the lifts are observed effects without significance testing.
minor comments (5)
  1. [§4.1.1] The 'leave-one-out strategy' is ambiguous. Specify which interaction is held out for train/validation/test (e.g., last interaction per user) and how the split handles users with very short histories.
  2. [§3.1, Eq. (4)] The attention equation would be clearer if Q, K, V were explicitly defined as L(e)W_Q, L(e)W_K, L(e)W_V. As written, the softmax argument and multiplication with V are understandable but notationally inconsistent with the surrounding text.
  3. [§3.3, Eq. (13)] Average pooling over historical items does not specify how padding or variable-length sequences are handled. A masking note would prevent ambiguity.
  4. [Table 1] The 'Sparsity' column is not defined. State the formula (e.g., 1 - #Interactions/(#User*#Item)) for clarity.
  5. [References] References [25] and [26] appear to describe the same work (SIM / Search-based User Interest Modeling). Please consolidate to avoid duplication.

Circularity Check

0 steps flagged

No significant circularity: MARS's retrieval-augmentation pipeline is an empirical method whose downstream CTR gains are measured independently of its training objectives.

full rationale

The paper's derivation chain is linear and self-contained: (i) item text/image embeddings are aligned with a Stein-kernel objective plus CTR loss (Eqs. 5-14); (ii) user embeddings are average-pooled item embeddings (Eq. 13); (iii) a high-active user is retrieved by cosine similarity (Eq. 15) and items are filtered by item-user similarity (Eqs. 16-17); (iv) the filtered items are prepended to the low-active user's sequence (Eq. 18) and a DIN model is trained and evaluated on AUC (Eq. 19). None of these equations defines the predicted CTR in terms of the fitted values used to generate the augmentation, nor is any target quantity constructed from the quantity it purports to predict. The alignment loss optimizes a kernel-based objective, but the scientific claim that this helps CTR is tested by held-out AUC, which is an external benchmark, not a restatement of the loss. Self-citations to TWIN/TWIN-V2 are contextual background on long-sequence modeling and are not load-bearing for MARS's retrieval mechanism; no uniqueness theorem is imported. The leave-one-out split is under-specified in the paper, raising a possible temporal-leakage concern about whether retrieved histories contain the target user's test item; however, that is an evaluation-validity risk, not a circularity by construction, and the paper does not reduce the prediction equation to the retrieval equation. Therefore no circular step meets the bar of exhibiting a specific reduction.

Axiom & Free-Parameter Ledger

4 free parameters · 5 axioms · 0 invented entities

The framework's success depends on the validity of the learned embedding space for retrieval, which is assumed and only tested empirically. Several hyperparameters (lambda, beta, theta, the 30/30 split) are chosen by hand or validation and are not reported, which limits the reader's ability to assess how robust the gains are.

free parameters (4)
  • lambda (alignment-entropy balance) = not reported
    Eq. (8) balances alignment fidelity vs. entropy regularization; set by validation, no sensitivity analysis.
  • beta (CTR-alignment balance) = not reported
    Eq. (14) weights the Stein loss against the CTR loss; tuned on validation.
  • theta (filtering threshold) = not reported
    Eq. (17) determines which retrieved items are kept; presumably tuned, but never specified.
  • low/high-active split (30%/30%) = 30% / 30%
    Chosen to define the target population; changes the difficulty of the task and likely the reported gains.
axioms (5)
  • domain assumption RBF kernel with median-distance bandwidth is a valid cross-modal similarity measure
    Eq. (5) uses a standard RBF kernel and calls the approach 'Stein-based', but no Stein kernel identity or unbiasedness is proven.
  • domain assumption Stein gradient estimator provides a tractable and correct score function approximation
    Eq. (7) relies on the estimator to compute the entropy gradient; the estimator itself is not defined in the paper.
  • domain assumption Average pooling of item embeddings yields a user embedding suitable for similarity search
    Eq. (13) averages fused item embeddings; no justification that this captures user interest distribution.
  • domain assumption Cosine similarity in the aligned multimodal space reflects interest compatibility
    Eq. (15) retrieves the nearest high-active user by cosine similarity; this is the core retrieval premise, asserted without proof.
  • domain assumption Prepended retrieved sequences do not introduce label shift or distribution mismatch
    Section 3.4 simply concatenates borrowed items; the paper does not analyze whether this changes the label distribution for the CTR task.

reviewed 2026-08-05 · how reviews work

0 comments
Cite this review

Pith. "Pith review of MARS: Modality-Aligned Retrieval for Sequence Augmented CTR Prediction." pith.science (2026). https://pith.science/paper/YCP6BX7D

@misc{pith2026250901184,
  author       = {Pith},
  title        = {Pith review of: MARS: Modality-Aligned Retrieval for Sequence Augmented CTR Prediction},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YCP6BX7D}},
  note         = {Machine review of arXiv:2509.01184}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Click-through rate (CTR) prediction serves as a cornerstone of recommender systems. Despite the strong performance of current CTR models based on user behavior modeling, they are still severely limited by interaction sparsity, especially in low-active user scenarios. To address this issue, data augmentation of user behavior is a promising research direction. However, existing data augmentation methods heavily rely on collaborative signals while overlooking the rich multimodal features of items, leading to insufficient modeling of low-active users. To alleviate this problem, we propose a novel framework \textbf{MARS} (\textbf{M}odality-\textbf{A}ligned \textbf{R}etrieval for \textbf{S}equence Augmented CTR Prediction). MARS utilizes a Stein kernel-based approach to align text and image features into a unified and unbiased semantic space to construct multimodal user embeddings. Subsequently, each low-active user's behavior sequence is augmented by retrieving, filtering, and concentrating the most similar behavior sequence of high-active users via multimodal user embeddings. Validated by extensive offline experiments and online A/B tests, our framework MARS consistently outperforms state-of-the-art baselines and achieves substantial growth on core business metrics within Kuaishou~\footnote{https://www.kuaishou.com/}. Consequently, MARS has been successfully deployed, serving the main traffic for hundreds of millions of users. To ensure reproducibility, we provide anonymous access to the implementation code~\footnote{https://github.com/wangshukuan/MARS}.

Figures

Figures reproduced from arXiv: 2509.01184 by Binhao Wang, Chao Feng, Fuzhen Zhuang, Shanqi Liu, Shukuan Wang, Xiang Li, Yanze Zhang, Yutian Xiao, Zhao Zhang.

Figure 1
Figure 1. Figure 1: (a) Evolution of CTR model sequence-length ca￾pacity; (b) Cumulative distribution of user sequence lengths over a one-month period on Kuaishou approximately 35% of users under 1,000 interactions. 2 Preliminaries In this section, we will briefly introduce the problem definition used in this paper, in order to aid in the reader’s understanding. 2.1 Problem Formulation CTR prediction is formulated as a superv… view at source ↗
Figure 2
Figure 2. Figure 2: An overview illustration of the MARS architecture. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: The deployment of MARS at Kuaishou. 3.5 Deployment of MARS We divide our hands-on practices for deploying MARS into three parts: Model Training, Online Service and Offline Runner, as shown in [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Impact of different padding strategies on model performance (AUC). [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗

discussion (0)

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

Forward citations

Cited by 1 Pith paper

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

  1. ScaleToT: Generalizing Structured LLM Reasoning for Billion-Scale Low-Activity User Modeling

    cs.AI 2026-06 unverdicted novelty 5.0

    ScaleToT builds typed user-state chains with entropy-guided Tree-of-Thought on a small LLM subset, trains a student via SFT and OSIPO, transfers representations to a profile encoder, and reports a 6.738% LT30 lift in ...

Reference graph

Works this paper leans on

47 extracted references · 21 canonical work pages · cited by 1 Pith paper · 4 internal anchors

  1. [1]

    Shuai Bai, Keqin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, Sibo Song, Kai Dang, Peng Wang, Shijie Wang, Jun Tang, Humen Zhong, Yuanzhi Zhu, Mingkun Yang, Zhaohai Li, Jianqiang Wan, Pengfei Wang, Wei Ding, Zheren Fu, Yiheng Xu, Jiabo Ye, Xi Zhang, Tianbao Xie, Zesen Cheng, Hang Zhang, Zhibo Yang, Haiyang Xu, and Junyang Lin. 2025. Qwen2.5-VL Technical Rep...

    Pith/arXiv arXiv 2025

  2. [2]

    Shuqing Bian, Wayne Xin Zhao, Jinpeng Wang, and Ji-Rong Wen. 2022. A relevant and diverse retrieval-enhanced data augmentation framework for sequential recommendation. In Proceedings of the 31st ACM International Conference on Information & Knowledge Management . 2923–2932

    work page 2022

  3. [3]

    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. arXiv:2205.10249 [cs.IR] https://arxiv.org/abs/ 2205.10249

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  4. [4]

    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

  5. [5]

    Junxuan Chen, Baigui Sun, Hao Li, Hongtao Lu, and Xian-Sheng Hua. 2016. Deep ctr prediction in display advertising. In Proceedings of the 24th ACM international conference on Multimedia. 811–820

    work page 2016

  6. [6]

    Qiwei Chen, Changhua Pei, Shanshan Lv, Chao Li, Junfeng Ge, and Wenwu Ou. 2021. End-to-End User Behavior Retrieval in Click-Through RatePrediction Model. arXiv:2108.04468 [cs.IR] https://arxiv.org/abs/2108.04468

    Pith/arXiv arXiv 2021

  7. [7]

    Qiwei Chen, Huan Zhao, Wei Li, Pipei Huang, and Wenwu Ou. 2019. Behavior sequence transformer for e-commerce recommendation in alibaba. InProceedings of the 1st international workshop on deep learning practice for high-dimensional sparse data. 1–4

    2019

  8. [8]

    Heng-Tze Cheng, Levent Koc, Jeremiah Harmsen, Tal Shaked, Tushar Chandra, Hrishi Aradhye, Glen Anderson, Greg Corrado, Wei Chai, Mustafa Ispir, et al

  9. [9]

    Paul Covington, Jay Adams, and Emre Sargin. 2016. Deep neural networks for youtube recommendations. In Proceedings of the 10th ACM conference on recommender systems. 191–198

    2016

  10. [10]

    Yizhou Dang, Yuting Liu, Enneng Yang, Guibing Guo, Linying Jiang, Xingwei Wang, and Jianzhe Zhao. 2024. Repeated Padding for Sequential Recommendation. In Proceedings of the 18th ACM Conference on Recommender Systems . 497–506

    work page 2024

  11. [11]

    Yizhou Dang, Yuting Liu, Enneng Yang, Minhan Huang, Guibing Guo, Jianzhe Zhao, and Xingwei Wang. 2025. Data Augmentation as Free Lunch: Exploring the Test-Time Augmentation for Sequential Recommendation. arXiv:2504.04843 [cs.IR] https://arxiv.org/abs/2504.04843

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  12. [12]

    Yizhou Dang, Jiahui Zhang, Yuting Liu, Enneng Yang, Yuliang Liang, Guibing Guo, Jianzhe Zhao, and Xingwei Wang. 2024. Augmenting Sequential Recommendation with Balanced Relevance and Diversity. arXiv:2412.08300 [cs.IR] https://arxiv. org/abs/2412.08300

    Pith/arXiv arXiv 2024

  13. [13]

    Yufei Feng, Fuyu Lv, Weichen Shen, Menghan Wang, Fei Sun, Yu Zhu, and Keping Yang. 2019. Deep session interest network for click-through rate prediction. arXiv preprint arXiv:1905.06482 (2019)

    Pith/arXiv arXiv 2019

  14. [14]

    Jingyue Gao, Shuguang Han, Han Zhu, Siran Yang, Yuning Jiang, Jian Xu, and Bo Zheng. 2023. Rec4ad: A free lunch to mitigate sample selection bias for ads ctr prediction in taobao. In Proceedings of the 32nd ACM International Conference on Information and Knowledge Management . 4574–4580

    work page 2023

  15. [15]

    Huifeng Guo, Ruiming Tang, Yunming Ye, Zhenguo Li, and Xiuqiang He. 2017. DeepFM: a factorization-machine based neural network for CTR prediction.arXiv preprint arXiv:1703.04247 (2017)

    Pith/arXiv arXiv 2017

  16. [16]

    Yuchin Juan, Yong Zhuang, Wei-Sheng Chin, and Chih-Jen Lin. 2016. Field- aware factorization machines for CTR prediction. In Proceedings of the 10th ACM conference on recommender systems . 43–50

    work page 2016

  17. [18]

    Qi Liu, Xuyang Hou, Haoran Jin, Zhe Wang, Defu Lian, Tan Qu, Jia Cheng, Jun Lei, et al. 2023. Deep Group Interest Modeling of Full Lifelong User Behaviors for CTR Prediction. arXiv preprint arXiv:2311.10764 (2023)

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  18. [19]

    Qiang Liu and Dilin Wang. 2019. Stein Variational Gradient Descent: A General Purpose Bayesian Inference Algorithm. arXiv:1608.04471 [stat.ML] https://arxiv. org/abs/1608.04471

    Pith/arXiv arXiv 2019

  19. [20]

    Qidong Liu, Fan Yan, Xiangyu Zhao, Zhaocheng Du, Huifeng Guo, Ruiming Tang, and Feng Tian. 2023. Diffusion augmentation for sequential recommenda- tion. In Proceedings of the 32nd ACM International conference on information and knowledge management. 1576–1586

    work page 2023

  20. [21]

    Yu, Julian McAuley, and Caiming Xiong

    Zhiwei Liu, Yongjun Chen, Jia Li, Philip S. Yu, Julian McAuley, and Caiming Xiong. 2021. Contrastive Self-supervised Sequential Recommendation with Robust Augmentation. arXiv:2108.06479 [cs.IR] https://arxiv.org/abs/2108.06479

    Pith/arXiv arXiv 2021

  21. [22]

    Ilya Loshchilov and Frank Hutter. 2017. Decoupled weight decay regularization. arXiv preprint arXiv:1711.05101 (2017)

    Pith/arXiv arXiv 2017

  22. [23]

    H Brendan McMahan, Gary Holt, David Sculley, Michael Young, Dietmar Ebner, Julian Grady, Lan Nie, Todd Phillips, Eugene Davydov, Daniel Golovin, et al

  23. [24]

    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 Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery &; Data Mining (KDD ’19) . ACM, 2671–2679. doi:10.1145/3292500. 3330666

    doi:10.1145/3292500 2019

  24. [25]

    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. In Proceedings of the 29th ACM International Conference on Information & Knowledge Management . 2685–2692

    2020

  25. [26]

    Pi Qi, Xiaoqiang Zhu, Guorui Zhou, Yujing Zhang, Zhe Wang, Lejian Ren, Ying Fan, and Kun Gai. 2020. Search-based User Interest Modeling with Lifelong Sequential Behavior Data for Click-Through Rate Prediction. arXiv:2006.05639 [cs.IR] https://arxiv.org/abs/2006.05639

    Pith/arXiv arXiv 2020

  26. [27]

    Jiarui Qin, Weinan Zhang, Rong Su, Zhirong Liu, Weiwen Liu, Guangpeng Zhao, Hao Li, Ruiming Tang, Xiuqiang He, and Yong Yu. 2023. Learning to retrieve user behaviors for click-through rate estimation. ACM Transactions on Information Systems 41, 4 (2023), 1–31

    work page 2023

  27. [28]

    Jiarui Qin, Weinan Zhang, Xin Wu, Jiarui Jin, Yuchen Fang, and Yong Yu. 2020. User Behavior Retrieval for Click-Through Rate Prediction. In Proceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR ’20) . ACM, 2347–2356. doi:10.1145/3397271.3401440

    work page arXiv 2020

  28. [29]

    Ruihong Qiu, Zi Huang, Hongzhi Yin, and Zijian Wang. 2022. Contrastive Learn- ing for Representation Degeneration Problem in Sequential Recommendation. In Proceedings of the Fifteenth ACM International Conference on Web Search and Data Mining (WSDM ’22) . ACM, 813–823. doi:10.1145/3488560.3498433

    arXiv 2022

  29. [30]

    Kan Ren, Jiarui Qin, Yuchen Fang, Weinan Zhang, Lei Zheng, Weijie Bian, Guorui Zhou, Jian Xu, Yong Yu, Xiaoqiang Zhu, and Kun Gai. 2019. Lifelong Sequential Modeling with Personalized Memorization for User Response Prediction. In Proceedings of the 42nd International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR ’19) . A...

    arXiv 2019

  30. [31]

    Steffen Rendle. 2010. Factorization machines. In2010 IEEE International conference on data mining. IEEE, 995–1000

    2010

  31. [32]

    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. In Proceedings of the 33rd ACM International Conference on Information and Knowledge Management. 4890–4897

    work page 2024

  32. [33]

    Weiping Song, Chence Shi, Zhiping Xiao, Zhijian Duan, Yewen Xu, Ming Zhang, and Jian Tang. 2019. Autoint: Automatic feature interaction learning via self- attentive neural networks. InProceedings of the 28th ACM international conference on information and knowledge management . 1161–1170

    work page 2019

  33. [34]

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

  34. [35]

    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. arXiv preprint arXiv:2302.13971 (2023)

    Pith/arXiv arXiv 2023

  35. [36]

    Hao Wang, Naiyan Wang, and Dit-Yan Yeung. 2015. Collaborative deep learning for recommender systems. In Proceedings of the 21th ACM SIGKDD international conference on knowledge discovery and data mining . 1235–1244

    work page 2015

  36. [37]

    Ruoxi Wang, Bin Fu, Gang Fu, and Mingliang Wang. 2017. Deep & cross network for ad click predictions. In Proceedings of the ADKDD’17 . 1–7

    2017

  37. [38]

    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. In Proceedings of the web conference 2021 . 1785–1797

    2021

  38. [39]

    Liangjian Wen, Xiasi Wang, Jianzhuang Liu, and Zenglin Xu. 2024. MVEB: Self-Supervised Learning with Multi-View Entropy Bottleneck. arXiv:2403.19078 [cs.CV] https://arxiv.org/abs/2403.19078

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  39. [40]

    Chuhan Wu, Fangzhao Wu, Lingjuan Lyu, Yongfeng Huang, and Xing Xie. 2022. FedCTR: Federated native ad CTR prediction with cross-platform user behavior data. ACM Transactions on Intelligent Systems and Technology (TIST) 13, 4 (2022), 1–19

    work page 2022

  40. [41]

    Xu Xie, Fei Sun, Zhaoyang Liu, Shiwen Wu, Jinyang Gao, Jiandong Zhang, Bolin Ding, and Bin Cui. 2022. Contrastive learning for sequential recommendation. In 2022 IEEE 38th international conference on data engineering (ICDE) . IEEE, 1259– 1273. KDD ’25, August 3–7, 2025, Toronto, ON, Canada Yutian Xiao, Shukuan Wang, Binhao Wang, Zhao Zhang, Yanze Zhang, S...

    work page 2022

  41. [42]

    Yanwu Yang and Panyu Zhai. 2022. Click-through rate prediction in online advertising: A literature review. Information Processing & Management 59, 2 (2022), 102853

    work page 2022

  42. [43]

    Yuren Zhang, Enhong Chen, Binbin Jin, Hao Wang, Min Hou, Wei Huang, and Runlong Yu. 2022. Clustering based behavior sampling with long sequential data for CTR prediction. In Proceedings of the 45th International ACM SIGIR Conference on Research and Development in Information Retrieval . 2195–2200

    work page 2022

  43. [44]

    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 , Vol. 33. 5941–5948

    2019

  44. [45]

    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 Proceedings of the 24th ACM SIGKDD international conference on knowledge discovery & data mining . 1059–1068. Received 20 February 2007; revised 12 March 2009; accepted 5 June 2009

    work page 2018

  45. [2013]

    In Proceedings of the 19th ACM SIGKDD international conference on Knowledge discovery and data mining

    Ad click prediction: a view from the trenches. In Proceedings of the 19th ACM SIGKDD international conference on Knowledge discovery and data mining . 1222–1230

    work page

  46. [2016]

    In Proceedings of the 1st workshop on deep learning for recommender systems

    Wide & deep learning for recommender systems. In Proceedings of the 1st workshop on deep learning for recommender systems . 7–10

  47. [2019]

    arXiv:1904.06690 [cs.IR] https://arxiv.org/abs/ 1904.06690

    BERT4Rec: Sequential Recommendation with Bidirectional Encoder Rep- resentations from Transformer. arXiv:1904.06690 [cs.IR] https://arxiv.org/abs/ 1904.06690

    Pith/arXiv arXiv 1904

This paper was first reviewed by deepseek-v4-flash on August 5, 2026.