pith. sign in

arxiv: 2605.15905 · v1 · pith:UO7YVABAnew · submitted 2026-05-15 · 💻 cs.IR · cs.AI

Generative Long-term User Interest Modeling for Click-Through Rate Prediction

Jiangli Shao , Kaifu Zheng , Hao Fang , Huimu Ye , Zhiwei Liu , Bo Zhang , Shu Han , Xingxing Wang This is my paper

Pith reviewed 2026-05-19 22:16 UTC · model grok-4.3

classification 💻 cs.IR cs.AI
keywords user interest modelingCTR predictiongenerative modelinglong-term behaviorsrecommendation systemsbehavior retrievalinterest fusiongating mechanisms
0
0 comments X

The pith

A generative module produces multiple target-independent interest distributions to capture diverse long-term user behaviors for click-through rate prediction.

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

The paper aims to show that long-term user interests in recommendation systems can be modeled more completely by generating several interest distributions first, without reference to the current target item. This generation step incorporates interactions among past behaviors and avoids the bias that comes from always centering retrieval on the target. A subsequent simple lookup then pulls relevant behaviors, after which gating combines the pieces into features for CTR prediction. If the generation step succeeds, systems gain both more varied interest signals and lower computational cost as behavior histories grow. The approach targets the common two-stage retrieval-plus-fusion pipeline used in advertising and recommendation.

Core claim

GenLI consists of an interest generation module that produces multiple target-independent interest distributions incorporating behavior interactions, a behavior retrieval module that selects related behaviors through constant-time lookup, and an interest fusion module that applies gating to form the final interest features for CTR prediction.

What carries the argument

The interest generation module, which creates multiple distributions representing different latent aspects of user interests without depending on the target item.

If this is right

  • Interest features become more complete because multiple distributions are generated instead of a single target-focused selection.
  • Retrieval cost drops to constant time per behavior, allowing the system to scale with growing user histories.
  • Diversity of represented interests increases, reducing the chance that secondary user preferences are ignored during prediction.
  • The overall pipeline achieves a tighter accuracy-efficiency trade-off for real-time serving in advertising systems.

Where Pith is reading between the lines

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

  • The same generative step could be applied to other sequential prediction tasks where exhaustive matching over histories becomes prohibitive.
  • Because distributions are produced without the target, the model might support faster pre-computation of user representations for batch serving.
  • If the generated distributions prove stable across sessions, they could serve as compact user embeddings for downstream tasks such as churn prediction.

Load-bearing premise

The generated interest distributions still reflect real latent user interests even when produced without any information about the target item.

What would settle it

An offline experiment on a large-scale CTR dataset in which replacing the generative module with a conventional target-centered retriever yields equal or higher AUC while maintaining the same online latency.

Figures

Figures reproduced from arXiv: 2605.15905 by Bo Zhang, Hao Fang, Huimu Ye, Jiangli Shao, Kaifu Zheng, Shu Han, Xingxing Wang, Zhiwei Liu.

Figure 1
Figure 1. Figure 1: The framework of GenLI. GenLI consists of an interest generation module (IGM), a behavior retrieval module (BRM), [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Inference time comparison of models. To fully understand the impact of generation methods on per￾formance, we conduct experiments on core parts of the interest generation module [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Performance of GenLI when taking different [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Accordingly, for all three models, AUC first increases with [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Performance of GenLI under different generated [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
read the original abstract

Modeling long-term user interests with massive historical user behaviors enhances click-through rate (CTR) prediction performance in advertising and recommendation systems. Typically, a two-stage framework is widely adopted, where a general search unit (GSU) first retrieves top-$k$ relevant behaviors towards the target item, and an exact search unit (ESU) generates interest features via tailored attention. However, current target-centered GSU would ignore other latent user interests, leading to incomplete and biased interest features. Additionally, the matching-based retrieval process in GSUs depends on the pairwise similarity score between target item and each historical behavior, which not only becomes time-consuming for online services as user behaviors continue to grow, but also overlooks the interaction information among user behaviors. To combat these problems, we propose a \textbf{Gen}erative \textbf{L}ong-term user \textbf{I}nterest model named GenLI for CTR prediction. GenLI consists of an interest generation module (IGM), a behavior retrieval module (BRM), and an interest fusion module (IFM). The IGM generates multiple interest distributions to indicate different aspects of real-time user interests, which is target-independent and incorporates interaction information among behaviors, ensuring complete and diverse interest features. The BRM selects related behaviors via a simple lookup operation, reducing the time complexity for weighting each behavior to $O(1)$. Finally, the IFM uses delicate gating mechanisms to generate interest features. Based on the generation process, GenLI improves the diversity of user interests and avoids complex matching-based behavioral retrieval, achieving a better balance between accuracy and efficiency for CTR prediction.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The manuscript proposes GenLI, a generative model for long-term user interest modeling in CTR prediction. It consists of an Interest Generation Module (IGM) that produces multiple target-independent interest distributions incorporating behavior interactions, a Behavior Retrieval Module (BRM) that performs simple lookup operations to achieve O(1) selection instead of pairwise matching, and an Interest Fusion Module (IFM) that applies gating mechanisms to generate interest features. The central claim is that this framework increases the diversity of captured user interests, eliminates the computational burden of matching-based retrieval in growing behavior histories, and achieves a better accuracy-efficiency balance than traditional two-stage GSU-ESU approaches.

Significance. If the empirical results support the claims, GenLI could provide a scalable generative alternative to similarity-based retrieval for handling massive user behavior sequences in recommendation and advertising systems. The shift to target-independent generation of diverse interest distributions addresses a recognized limitation of target-centered methods and, if validated, may influence future work on efficient interest modeling.

major comments (2)
  1. Abstract (IGM description): The claim that the IGM generates distributions that 'faithfully represent multiple latent aspects of real user interests' while remaining target-independent is load-bearing for the accuracy component of the accuracy-efficiency balance. User interests in CTR tasks are typically target- and context-dependent; without any conditioning on the target item during generation, the distributions risk including irrelevant or outdated aspects that the subsequent BRM lookup (which lacks explicit similarity or interaction scoring) cannot filter, potentially degrading features relative to conventional GSU methods.
  2. Abstract (BRM description): The BRM is stated to select 'related behaviors via a simple lookup operation' with O(1) complexity. However, the mechanism determining which behaviors are sufficiently relevant for a given target—absent any pairwise scoring or target interaction—is not specified, leaving open whether the selected behaviors remain adequate for the IFM to produce accurate interest features.
minor comments (1)
  1. Abstract: The phrase 'delicate gating mechanisms' in the IFM description is vague; a brief indication of how these mechanisms differ from standard gating or attention would improve clarity.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. The comments highlight important aspects of the abstract's claims regarding the IGM and BRM. We address each point below with clarifications drawn from the full method description and indicate where revisions will be made to improve clarity.

read point-by-point responses

  1. Referee: Abstract (IGM description): The claim that the IGM generates distributions that 'faithfully represent multiple latent aspects of real user interests' while remaining target-independent is load-bearing for the accuracy component of the accuracy-efficiency balance. User interests in CTR tasks are typically target- and context-dependent; without any conditioning on the target item during generation, the distributions risk including irrelevant or outdated aspects that the subsequent BRM lookup (which lacks explicit similarity or interaction scoring) cannot filter, potentially degrading features relative to conventional GSU methods.

    Authors: We appreciate the referee's observation on the potential risks of target-independent generation. The IGM is designed to produce multiple distributions by modeling interactions across the full behavior sequence, explicitly to capture latent aspects that target-centered GSU methods often overlook or bias against. While generation itself does not condition on the target, the subsequent IFM employs gating mechanisms that incorporate target-item features to weigh and select from these distributions, thereby mitigating inclusion of irrelevant aspects during feature fusion. This separation enables the diversity benefit while preserving relevance. We agree the abstract phrasing could better highlight the IFM's filtering role and have revised it accordingly, along with a short clarifying paragraph in Section 3.1. revision: partial

  2. Referee: Abstract (BRM description): The BRM is stated to select 'related behaviors via a simple lookup operation' with O(1) complexity. However, the mechanism determining which behaviors are sufficiently relevant for a given target—absent any pairwise scoring or target interaction—is not specified, leaving open whether the selected behaviors remain adequate for the IFM to produce accurate interest features.

    Authors: The BRM achieves O(1) lookup by associating each historical behavior with the interest distributions generated by the IGM during the offline or pre-computation stage; at inference, behaviors are retrieved directly via these pre-assigned distribution indices rather than computing pairwise similarities with the target. This mechanism is described in detail in Section 3.2, where we explain the assignment process based on behavior-distribution affinity scores computed once per user history. We acknowledge that the abstract omitted this key detail for brevity and have expanded the BRM description in the revised abstract to include a concise statement of the lookup basis. revision: yes

Circularity Check

0 steps flagged

No circularity: architecture proposal is self-contained without reducing claims to fitted inputs or self-citations

full rationale

The paper proposes GenLI consisting of IGM (target-independent interest generation), BRM (simple lookup retrieval), and IFM (gating fusion). Claims about diversity, completeness, and accuracy-efficiency balance are presented as direct consequences of these design choices rather than derived via equations that equate outputs to inputs by construction. No fitting procedures, self-definitional loops, or load-bearing self-citations appear in the provided description. The derivation chain remains independent of the target result and does not rename known patterns or smuggle ansatzes.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The model rests on the domain assumption that user behavior sequences contain multiple separable latent interest aspects that can be generated without target information; no free parameters or invented physical entities are specified in the abstract.

axioms (1)
  • domain assumption User behavior histories contain multiple latent interest aspects that can be represented as target-independent distributions.
    Invoked in the description of the interest generation module (IGM).
invented entities (1)
  • Interest generation module (IGM) no independent evidence
    purpose: Generates multiple interest distributions to capture diverse latent user interests.
    New architectural component introduced to replace target-centered retrieval.

pith-pipeline@v0.9.0 · 5839 in / 1316 out tokens · 54788 ms · 2026-05-19T22:16:37.944334+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

43 extracted references · 43 canonical work pages · 3 internal anchors

  1. [1]

    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. ACM, 2974–2983

    work page 2022

  2. [2]

    Jianxin Chang, Chenbin Zhang, Zhiyi Fu, Xiaoxue Zang, Lin Guan, Jing Lu, Yiqun Hui, Dewei Leng, Yanan Niu, Yang Song, and Kun Gai. 2023. TWIN: TWo-stage Interest Network for Lifelong User Behavior Modeling in CTR Prediction at Kuaishou. InProceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, KDD 2023. ACM, 3785–3794

    work page 2023

  3. [3]

    Junxuan Chen, Baigui Sun, Hao Li, Hongtao Lu, and Xian-Sheng Hua. 2016. Deep CTR Prediction in Display Advertising. InProceedings of the 24th ACM International Conference on Multimedia(Amsterdam, The Netherlands)(MM ’16). 811–820

    work page 2016

  4. [4]

    Qiwei Chen, Changhua Pei, Shanshan Lv, Chao Li, Junfeng Ge, and Wenwu Ou

    work page

  5. [5]

    End-to-end user behavior retrieval in click-through rateprediction model.arXiv preprint arXiv:2108.04468, 2021

    End-to-End User Behavior Retrieval in Click-Through Rate Prediction Model.CoRRabs/2108.04468 (2021). arXiv:2108.04468

    work page arXiv 2021

  6. [6]

    Heng-Tze Cheng, Levent Koc, Jeremiah Harmsen, Tal Shaked, Tushar Chandra, Hrishi Aradhye, Glen Anderson, Greg Corrado, Wei Chai, Mustafa Ispir, Rohan Anil, Zakaria Haque, Lichan Hong, Vihan Jain, Xiaobing Liu, and Hemal Shah

    work page

  7. [7]

    InProceedings of the 1st Workshop on Deep Learning for Recommender Systems, DLRS (RecSys), 2016

    Wide & Deep Learning for Recommender Systems. InProceedings of the 1st Workshop on Deep Learning for Recommender Systems, DLRS (RecSys), 2016. ACM, 7–10

    work page 2016

  8. [8]

    Yashar Deldjoo, Zhankui He, Julian McAuley, Anton Korikov, Scott Sanner, Arnau Ramisa, René Vidal, Maheswaran Sathiamoorthy, Atoosa Kasirzadeh, and Silvia Milano. 2024. A Review of Modern Recommender Systems Using Generative Models (Gen-RecSys).arXiv preprint arXiv:2404.00579(2024)

    work page arXiv 2024

  9. [9]

    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. InProceedings of the Twenty-Eighth International Joint Conference on Artificial Intelligence, IJCAI 2019. 2301–2307

    work page 2019

  10. [10]

    Liqiong Gu. 2021. Ad Click-Through Rate Prediction: A Survey. InDatabase Systems for Advanced Applications. DASFAA 2021 International Workshops. Cham, 140–153

    work page 2021

  11. [11]

    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)

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  12. [12]

    Jinming Li, Wentao Zhang, Tian Wang, Guanglei Xiong, Alan Lu, and Gerard Medioni. 2023. GPT4Rec: A Generative Framework for Personalized Recommen- dation and User Interests Interpretation. InProceedings of the 2023 SIGIR Workshop on eCommerce co-located with the 46th International ACM SIGIR Conference on Research and Development in Information Retrieval (...

    work page 2023

  13. [13]

    Xiaoxi Li, Yujia Zhou, and Zhicheng Dou. 2024. UniGen: A Unified Generative Framework for Retrieval and Question Answering with Large Language Models. InThirty-Eighth AAAI Conference on Artificial Intelligence, AAAI 2024. AAAI Press, 8688–8696

    work page 2024

  14. [14]

    Yongqi Li, Nan Yang, Liang Wang, Furu Wei, and Wenjie Li. 2024. Learning to Rank in Generative Retrieval. InThirty-Eighth AAAI Conference on Artificial Intelligence, AAAI 2024. AAAI Press, 8716–8723

    work page 2024

  15. [15]

    Qi Liu, Xuyang Hou, Haoran Jin, Jin Chen, Zhe Wang, Defu Lian, Tan Qu, Jia Cheng, and Jun Lei. 2023. Deep Group Interest Modeling of Full Lifelong User Behaviors for CTR Prediction.CoRRabs/2311.10764 (2023). doi:10.48550/ARXIV. 2311.10764 arXiv:2311.10764

    work page internal anchor Pith review doi:10.48550/arxiv 2023

  16. [16]

    Qi Liu, Xuyang Hou, Defu Lian, Zhe Wang, Haoran Jin, Jia Cheng, and Jun Lei. 2024. AT4CTR: Auxiliary Match Tasks for Enhancing Click-Through Rate Prediction. InThirty-Eighth AAAI Conference on Artificial Intelligence, AAAI 2024. AAAI Press, 8787–8795

    work page 2024

  17. [17]

    Shuchang Liu, Qingpeng Cai, Zhankui He, Bowen Sun, Julian McAuley, Dong Zheng, Peng Jiang, and Kun Gai. 2023. Generative Flow Network for Listwise Rec- ommendation. InProceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining(Long Beach, CA, USA)(KDD ’23). 1524–1534. Conference acronym ’XX, June 03–05, 2018, Woodstock, NY et al

    work page 2023

  18. [18]

    Xinwei Long, Jiali Zeng, Fandong Meng, Zhiyuan Ma, Kaiyan Zhang, Bowen Zhou, and Jie Zhou. 2024. Generative Multi-Modal Knowledge Retrieval with Large Language Models. InThirty-Eighth AAAI Conference on Artificial Intelligence, AAAI 2024. AAAI Press, 18733–18741

    work page 2024

  19. [19]

    Jianmo Ni, Jiacheng Li, and Julian McAuley. 2019. Justifying Recommendations using Distantly-Labeled Reviews and Fine-Grained Aspects. InProceedings of the 2019 Conference on Empirical Methods in Natural Language Processing (EMNLP). Hong Kong, China, 188–197

    work page 2019

  20. [20]

    Aleksandr V Petrov and Craig Macdonald. 2023. Generative sequential recom- mendation with gptrec.arXiv preprint arXiv:2306.11114(2023)

    work page arXiv 2023

  21. [21]

    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, KDD 2019. ACM, 2671–2679

    work page 2019

  22. [22]

    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. InCIKM ’20: The 29th ACM International Conference on Information and Knowledge Management. ACM, 2685–2692

    work page 2020

  23. [23]

    Tao Qi, Fangzhao Wu, Chuhan Wu, Peiru Yang, Yang Yu, Xing Xie, and Yongfeng Huang. 2021. HieRec: Hierarchical user interest modeling for personalized news recommendation.arXiv preprint arXiv:2106.04408(2021)

    work page arXiv 2021

  24. [24]

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

    work page 2020

  25. [25]

    Alec Radford, Karthik Narasimhan, Tim Salimans, Ilya Sutskever, et al . 2018. Improving language understanding by generative pre-training. (2018)

    work page 2018

  26. [26]

    Shashank Rajput, Nikhil Mehta, Anima Singh, Raghunandan Hulikal Keshavan, Trung Vu, Lukasz Heldt, Lichan Hong, Yi Tay, Vinh Tran, Jonah Samost, Maciej Kula, Ed Chi, and Maheswaran Sathiamoorthy. 2023. Recommender Systems with Generative Retrieval. InAdvances in Neural Information Processing Systems, Vol. 36. 10299–10315

    work page 2023

  27. [27]

    Stephen Robertson, Hugo Zaragoza, and Michael Taylor. 2004. Simple BM25 extension to multiple weighted fields. InProceedings of the Thirteenth ACM International Conference on Information and Knowledge Management (CIKM ’04). 42–49

    work page 2004

  28. [28]

    Caitlin Sadowski and Greg Levin. 2007. Simhash: Hash-based similarity detection

    work page 2007

  29. [29]

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Ł ukasz Kaiser, and Illia Polosukhin. 2017. Attention is All you Need. InAdvances in Neural Information Processing Systems, Vol. 30

    work page 2017

  30. [30]

    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 (WWW ’21). 1785–1797

    work page 2021

  31. [31]

    Zhibo Xiao, Luwei Yang, Wen Jiang, Yi Wei, Yi Hu, and Hao Wang. 2020. Deep Multi-Interest Network for Click-through Rate Prediction. InProceedings of the 29th ACM International Conference on Information & Knowledge Management (CIKM ’20). 2265–2268

    work page 2020

  32. [32]

    Weinan Xu, Hengxu He, Minshi Tan, Yunming Li, Jun Lang, and Dongbai Guo

    work page

  33. [33]

    InProceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval(Virtual Event, China)(SIGIR ’20)

    Deep Interest with Hierarchical Attention Network for Click-Through Rate Prediction. InProceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval(Virtual Event, China)(SIGIR ’20). 1905–1908

    work page 1905

  34. [34]

    Yanwu Yang and Panyu Zhai. 2022. Click-through rate prediction in online advertising: A literature review.Information Processing & Management59, 2 (2022), 102853

    work page 2022

  35. [35]

    Hongzhi Yin, Bin Cui, Ling Chen, Zhiting Hu, and Zi Huang. 2014. A temporal context-aware model for user behavior modeling in social media systems. In Proceedings of the 2014 ACM SIGMOD International Conference on Management of Data (SIGMOD ’14). 1543–1554

    work page 2014

  36. [36]

    Dani Yogatama, Chris Dyer, Wang Ling, and Phil Blunsom. 2017. Generative and discriminative text classification with recurrent neural networks.arXiv preprint arXiv:1703.01898(2017)

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  37. [37]

    Kaifu Zheng, Lu Wang, Yu Li, Xusong Chen, Hu Liu, Jing Lu, Xiwei Zhao, Chang- ping Peng, Zhangang Lin, and Jingping Shao. 2022. Implicit User Awareness Modeling via Candidate Items for CTR Prediction in Search Ads. InWWW ’22: The ACM Web Conference 2022. ACM, 246–255

    work page 2022

  38. [38]

    Chang Zhou, Jinze Bai, Junshuai Song, Xiaofei Liu, Zhengchao Zhao, Xiusi Chen, and Jun Gao. 2018. ATRank: An Attention-Based User Behavior Modeling Frame- work for Recommendation. InProceedings of the Thirty-Second AAAI Conference on Artificial Intelligence, (AAAI-18). AAAI Press, 4564–4571

    work page 2018

  39. [39]

    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. InThe Thirty-Third AAAI Conference on Artificial Intelligence, AAAI

    work page 2019

  40. [40]

    AAAI Press, 5941–5948

    work page

  41. [41]

    Guorui Zhou, Xiaoqiang Zhu, Chengru 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. InProceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, KDD 2018. ACM, 1059–1068

    work page 2018

  42. [42]

    Han Zhu, Xiang Li, Pengye Zhang, Guozheng Li, Jie He, Han Li, and Kun Gai

    work page

  43. [43]

    InProceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, KDD 2018

    Learning Tree-based Deep Model for Recommender Systems. InProceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, KDD 2018. ACM, 1079–1088. Received xx; revised xx; accepted xx

    work page 2018