arxiv: 2604.05314 · v1 · submitted 2026-04-07 · 💻 cs.IR

Recognition: 2 theorem links

· Lean Theorem

Next-Scale Generative Reranking: A Tree-based Generative Rerank Method at Meituan

Shuli Wang , Changhao Li , Ke Fan , Senjie Kou Junwei Yin , Chi Wang , Yinhua Zhu , Haitao Wang , Xingxing Wang

Authors on Pith no claims yet

Pith reviewed 2026-05-10 19:50 UTC · model grok-4.3

classification 💻 cs.IR

keywords generative rerankingtree-based generationmulti-scale modelingrecommendation systemsnext-scale generatormulti-scale evaluatorcontextual rerankingindustrial deployment

0 comments

The pith

A tree-based next-scale generator builds recommendation lists from coarse user interests to fine details while aligning training signals at every scale.

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

Generative rerankers in multi-stage recommendation systems often produce suboptimal lists because they lack either local item context or global list structure, and because the generator and evaluator pursue inconsistent goals during training. The paper introduces NSGR, a framework that grows lists progressively across scales in a tree structure so each step incorporates both perspectives. A multi-scale neighbor loss then delivers scale-specific guidance from a tree-based evaluator to keep the two components aligned. Experiments on public and industrial datasets plus live deployment on a food delivery platform support the approach. A sympathetic reader cares because improved reranking directly changes the ordered items users see and act on in high-volume choice environments.

Core claim

NSGR establishes that a next-scale generator which expands a recommendation list from user interests in a coarse-to-fine tree manner, guided by a multi-scale neighbor loss that supplies scale-specific signals from a tree-based multi-scale evaluator, overcomes the dual problems of missing local-global balance and generator-evaluator goal inconsistency that affect both autoregressive and non-autoregressive generative rerankers.

What carries the argument

The next-scale generator (NSG) that progressively expands lists scale by scale together with the multi-scale neighbor loss drawn from the tree-based multi-scale evaluator (MSE) at each level.

If this is right

The generator receives both local item interactions and global list coherence at every expansion step.
Training guidance becomes consistent across scales instead of conflicting between generator and evaluator.
The framework scales to industrial volumes as shown by live deployment on the Meituan food delivery platform.
Effectiveness holds across both public benchmarks and private industrial data.

Where Pith is reading between the lines

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

The same coarse-to-fine tree expansion could be tested on other list-generation tasks such as search result ordering or playlist creation where hierarchical context matters.
If the scales naturally reflect user interest hierarchies, the method might reduce the need for separate context-modeling layers in rerankers.
One could measure whether adding finer scales beyond the current design yields diminishing returns or requires more training data.

Load-bearing premise

The tree-based multi-scale structure and neighbor loss actually supply independent local and global signals without introducing new inconsistencies or requiring extensive post-hoc tuning.

What would settle it

An ablation that removes either the tree structure or the multi-scale neighbor loss and shows no measurable drop in ranking metrics such as NDCG or click-through rate on the same public and Meituan datasets would falsify the claim.

Figures

Figures reproduced from arXiv: 2604.05314 by Changhao Li, Chi Wang, Haitao Wang, Ke Fan, Senjie Kou Junwei Yin, Shuli Wang, Xingxing Wang, Yinhua Zhu.

**Figure 1.** Figure 1: Demo of NSGR. When generating a list of length 4, we gradually determine the position of each item from the mixed [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗

**Figure 2.** Figure 2: The overall architecture of NSGR. multi-scale Self-Attention (SA) [30]: e (1) 𝑡 = e1,𝑚 = SA(x 𝑠 1 ||x 𝑠 2 ||...||x 𝑠 𝑚), e (2) 𝑡 = e1,𝑚/2 = SA(x 𝑠 1 ||...||x 𝑠 𝑡 ||...||x 𝑠 𝑚/2 ), ... e (𝐾) 𝑡 = e𝑡,𝑡+1 = SA(x 𝑠 𝑡 ||x 𝑠 𝑡+1 ), (7) where 𝐾 = log2 𝑚. Note that the SA layer in the above formula does not have position encoding, which can increase reusability and reduce calculations. Then we collect multi-scale c… view at source ↗

**Figure 4.** Figure 4: Multi-scale neighbor loss demo. First, we collect neighbor lists at multiple scales. As illustrated in [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗

**Figure 3.** Figure 3: Next-scale generation process demo. 4.3 Training with Multi-Scale Neighbor Loss As mentioned before, the goal inconsistency between the evaluator and the generator complicates the transfer of guidance. We will introduce our solution in detail below. 4.3.1 Training of the MSE. The MSE is tuned to fit the listwise value of items. We train the MSE using real data collected from online logs, including exposur… view at source ↗

**Figure 5.** Figure 5: Performance comparison within the𝐴 8 8 permutation space distribution. As shown in [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗

**Figure 6.** Figure 6: Positional distribution of Top-10 items following NSGR. [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗

**Figure 7.** Figure 7: Architecture of the online deployment with NSGR. [PITH_FULL_IMAGE:figures/full_fig_p009_7.png] view at source ↗

read the original abstract

In modern multi-stage recommendation systems, reranking plays a critical role by modeling contextual information. Due to inherent challenges such as the combinatorial space complexity, an increasing number of methods adopt the generative paradigm: the generator produces the optimal list during inference, while an evaluator guides the generator's optimization during the training phase. However, these methods still face two problems. Firstly, these generators fail to produce optimal generation results due to the lack of both local and global perspectives, regardless of whether the generation strategy is autoregressive or non-autoregressive. Secondly, the goal inconsistency problem between the generator and the evaluator during training complicates the guidance signal and leading to suboptimal performance. To address these issues, we propose the \textbf{N}ext-\textbf{S}cale \textbf{G}eneration \textbf{R}eranking (NSGR), a tree-based generative framework. Specifically, we introduce a next-scale generator (NSG) that progressively expands a recommendation list from user interests in a coarse-to-fine manner, balancing global and local perspectives. Furthermore, we design a multi-scale neighbor loss, which leverages a tree-based multi-scale evaluator (MSE) to provide scale-specific guidance to the NSG at each scale. Extensive experiments on public and industrial datasets validate the effectiveness of NSGR. And NSGR has been successfully deployed on the Meituan food delivery platform.

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.

Desk Editor's Note private letter to a colleague

NSGR adds a tree-structured progressive generator and multi-scale neighbor loss to generative reranking to fix local/global imbalance and generator-evaluator mismatch, but the abstract supplies no numbers or ablations to show the fixes actually work.

read the letter

The core idea is a next-scale generator that grows a recommendation list scale by scale inside a tree, starting coarse and adding finer details, paired with a multi-scale evaluator that supplies per-scale guidance through a neighbor loss. This framing tries to give the generator both global structure and local accuracy while keeping the training signal consistent, which the authors say current autoregressive and non-autoregressive methods miss. The tree construction and scale-specific loss are the concrete new pieces relative to the generative rerankers cited in the abstract. Deployment at Meituan is also stated, which at least signals that someone ran it at production scale. That combination is worth noting for anyone already experimenting with generative reranking in large systems. The motivation section is straightforward and the problems it names are real in practice. The soft spots sit in the evidence. The abstract claims extensive experiments and successful deployment yet shows zero metrics, no baseline comparisons, no ablation on the number of scales or the neighbor loss contribution, and no error analysis. Without those, it is impossible to tell whether the tree traversal keeps the local and global signals independent or whether the loss terms create hidden coupling as the stress-test note worries. The free parameter of scale count is mentioned but not explored. This paper is mainly for engineers and researchers who already work on industrial reranking pipelines and want to see a new architectural sketch. A reader who needs concrete gains or reproducible details will not get much yet. I would send it to peer review because the claimed deployment at Meituan scale makes the tree-plus-multi-scale design worth checking in full, even though the current version will need heavy revision to include the missing results and ablations.

Referee Report

2 major / 2 minor

Summary. The manuscript proposes Next-Scale Generative Reranking (NSGR), a tree-based generative reranking framework for multi-stage recommendation systems. It identifies two limitations in prior generative rerankers: generators that lack simultaneous local and global perspectives (autoregressive or non-autoregressive) and goal inconsistency between generator and evaluator during training. NSGR introduces a Next-Scale Generator (NSG) that expands lists in a coarse-to-fine tree traversal from user interests and a multi-scale neighbor loss that employs a tree-based Multi-Scale Evaluator (MSE) to supply scale-specific guidance. The authors state that extensive experiments on public and industrial datasets confirm effectiveness and that NSGR has been deployed on the Meituan food delivery platform.

Significance. If the empirical gains and the claimed independence of local/global signals are substantiated, the work would offer a concrete architectural pattern for incorporating multi-scale structure into generative reranking, potentially improving both optimization stability and list quality in contextual recommendation. The reported production deployment on a large-scale platform constitutes a practical strength that is uncommon in purely academic IR papers and would strengthen the case for real-world impact.

major comments (2)

[Abstract] Abstract: the statement that 'extensive experiments validate the effectiveness' and that NSGR 'has been successfully deployed' is unsupported by any reported metrics, ablation tables, baseline comparisons, or error analysis, rendering the central effectiveness claim impossible to assess from the provided evidence.
[§4] §4 (multi-scale neighbor loss): the claim that the tree-based MSE supplies independent local and global signals while eliminating goal inconsistency rests on the unproven assumption that scale-specific loss terms are orthogonal and produce no cross-scale gradient conflicts; no derivation, orthogonality argument, or bounded-interference analysis is supplied, which is load-bearing for the resolution of both stated problems.

minor comments (2)

The first use of the acronyms NSG and MSE should be accompanied by explicit definitions rather than relying on the expanded forms alone.
The description of the tree traversal in the NSG would benefit from a small illustrative diagram or pseudocode to clarify the coarse-to-fine expansion process.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful reading and constructive comments on our manuscript. We address each major comment below and have revised the manuscript to incorporate the feedback where appropriate.

read point-by-point responses

Referee: [Abstract] Abstract: the statement that 'extensive experiments validate the effectiveness' and that NSGR 'has been successfully deployed' is unsupported by any reported metrics, ablation tables, baseline comparisons, or error analysis, rendering the central effectiveness claim impossible to assess from the provided evidence.

Authors: We agree that the abstract would benefit from greater specificity to allow readers to immediately assess the strength of the claims. The full manuscript contains detailed experimental results in Section 5 (including baseline comparisons, ablation studies, and error analysis on public and industrial datasets) and deployment details in Section 6. In the revised manuscript, we have updated the abstract to include key quantitative improvements (e.g., relative gains on public benchmarks and the industrial dataset) and a concise reference to the production deployment outcomes, while remaining within typical abstract length limits. revision: yes
Referee: [§4] §4 (multi-scale neighbor loss): the claim that the tree-based MSE supplies independent local and global signals while eliminating goal inconsistency rests on the unproven assumption that scale-specific loss terms are orthogonal and produce no cross-scale gradient conflicts; no derivation, orthogonality argument, or bounded-interference analysis is supplied, which is load-bearing for the resolution of both stated problems.

Authors: We acknowledge that the original manuscript does not contain a formal derivation proving orthogonality of the scale-specific loss terms or a bounded analysis of potential cross-scale gradient interference. The multi-scale neighbor loss is motivated by the hierarchical tree structure of NSG, where each scale operates on progressively refined candidate lists and the MSE provides guidance at the corresponding granularity; this design aims to align training signals without explicit cross-scale conflicts. We provide supporting empirical evidence via ablation studies showing stable convergence and additive gains from each scale term. In the revision, we have expanded §4 with a dedicated discussion paragraph explaining the rationale for scale independence based on the coarse-to-fine traversal and added further ablation results isolating the contribution of individual loss terms. A complete theoretical proof remains challenging given the non-convex optimization landscape, but the added material strengthens the justification for the claimed benefits. revision: partial

Circularity Check

0 steps flagged

NSGR presented as architectural proposal with no self-referential derivations or fitted predictions

full rationale

The paper proposes NSGR as a new tree-based generative reranking framework that uses a next-scale generator (NSG) for coarse-to-fine list expansion and a multi-scale neighbor loss with a tree-based multi-scale evaluator (MSE) to address local/global perspectives and generator-evaluator inconsistency. No equations, parameter-fitting procedures, or self-citations are described that reduce claimed improvements to quantities defined by the same evaluation data or by construction. The method is introduced at the level of architecture and loss design, with validation via experiments on public and industrial datasets plus real-world deployment. This qualifies as a self-contained proposal against external benchmarks, with no load-bearing steps that collapse to the inputs.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 2 invented entities

The central claim rests on the unproven premise that progressive tree expansion plus per-scale loss terms will simultaneously solve both the local-global perspective gap and the generator-evaluator goal inconsistency; no independent verification of these mechanisms is supplied beyond the abstract claim of deployment success.

free parameters (1)

number of scales
The depth of the coarse-to-fine tree is a design hyperparameter whose value is not reported.

axioms (1)

domain assumption Combinatorial space complexity prevents direct optimization of the full list
Invoked in the first sentence of the abstract as the reason generative methods are needed.

invented entities (2)

next-scale generator (NSG) no independent evidence
purpose: Progressively expands the recommendation list from user interests in coarse-to-fine manner
New architectural component introduced to balance global and local perspectives.
multi-scale evaluator (MSE) no independent evidence
purpose: Supplies scale-specific guidance via multi-scale neighbor loss
New component introduced to resolve goal inconsistency between generator and evaluator.

pith-pipeline@v0.9.0 · 5569 in / 1437 out tokens · 123073 ms · 2026-05-10T19:50:45.633350+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

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

IndisputableMonolith/Foundation/AlexanderDuality.lean alexander_duality_circle_linking (D=3 from 2^D=8) echoes

?

echoes
ECHOES: this paper passage has the same mathematical shape or conceptual pattern as the Recognition theorem, but is not a direct formal dependency.

next-scale generator (NSG) that progressively expands a recommendation list from user interests in a coarse-to-fine manner... 'one-generates-two, two-generate-four' pattern... K=log2 m steps
IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel (J-cost uniqueness) unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

multi-scale neighbor loss... scale-specific guidance to the NSG at each scale

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

44 extracted references · 13 canonical work pages · 2 internal anchors

[2]

Qingyao Ai, Keping Bi, Jiafeng Guo, and W Bruce Croft. 2018. Learning a deep listwise context model for ranking refinement. InThe 41st international ACM SIGIR conference on research & development in information retrieval. 135–144

2018
[3]

Areej Alsini, Du Q Huynh, and Amitava Datta. 2020. Hit ratio: An evaluation metric for hashtag recommendation.arXiv preprint arXiv:2010.01258(2020)

work page arXiv 2020
[4]

Irwan Bello, Sayali Kulkarni, Sagar Jain, Craig Boutilier, Ed Chi, Elad Eban, Xiyang Luo, Alan Mackey, and Ofer Meshi. 2018. Seq2Slate: Re-ranking and slate optimization with RNNs.arXiv preprint arXiv:1810.02019(2018)

work page arXiv 2018
[5]

Christopher JC Burges. 2010. From ranknet to lambdarank to lambdamart: An overview.Learning11, 23-581 (2010), 81

2010
[6]

Zhe Cao, Tao Qin, Tie-Yan Liu, Ming-Feng Tsai, and Hang Li. 2007. Learning to rank: from pairwise approach to listwise approach. InProceedings of the 24th international conference on Machine learning. 129–136

2007
[7]

Moses S Charikar. 2002. Similarity estimation techniques from rounding algo- rithms. InProceedings of the thiry-fourth annual ACM symposium on Theory of computing. 380–388

2002
[8]

Chi Chen, Hui Chen, Kangzhi Zhao, Junsheng Zhou, Li He, Hongbo Deng, Jian Xu, Bo Zheng, Yong Zhang, and Chunxiao Xing. 2022. EXTR: Click-Through Rate Prediction with Externalities in E-Commerce Sponsored Search. InProceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining. 2732–2740

2022
[9]

Qiwei Chen, Changhua Pei, Shanshan Lv, Chao Li, Junfeng Ge, and Wenwu Ou
[10]

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 rateprediction model. arXiv preprint arXiv:2108.04468(2021)

work page arXiv 2021
[11]

Heng-Tze Cheng, Levent Koc, Jeremiah Harmsen, Tal Shaked, Tushar Chandra, Hrishi Aradhye, Glen Anderson, Greg Corrado, Wei Chai, Mustafa Ispir, et al
[12]

InProceedings of the 1st workshop on deep learning for recommender systems

Wide & deep learning for recommender systems. InProceedings of the 1st workshop on deep learning for recommender systems. 7–10
[13]

Yufei Feng, Yu Gong, Fei Sun, Junfeng Ge, and Wenwu Ou. 2021. Revisit recom- mender system in the permutation prospective.arXiv preprint arXiv:2102.12057 (2021)

work page arXiv 2021
[14]

Yufei Feng, Binbin Hu, Yu Gong, Fei Sun, Qingwen Liu, and Wenwu Ou. 2021. GRN: Generative Rerank Network for Context-wise Recommendation.arXiv preprint arXiv:2104.00860(2021)

work page arXiv 2021
[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)

work page arXiv 2017
[16]

Yupeng Hou, Jiacheng Li, Ashley Shin, Jinsung Jeon, Abhishek Santhanam, Wei Shao, Kaveh Hassani, Ning Yao, and Julian McAuley. 2025. Generating long semantic IDs in parallel for recommendation. InProceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V. 2. 956–966

2025
[17]

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

2021
[18]

Ray Jiang, Sven Gowal, Timothy A Mann, and Danilo J Rezende. 2018. Beyond greedy ranking: Slate optimization via list-CVAE.arXiv preprint arXiv:1803.01682 (2018)

work page arXiv 2018
[19]

Jianxun Lian, Xiaohuan Zhou, Fuzheng Zhang, Zhongxia Chen, Xing Xie, and Guangzhong Sun. 2018. xdeepfm: Combining explicit and implicit feature in- teractions for recommender systems. InProceedings of the 24th ACM SIGKDD international conference on knowledge discovery & data mining. 1754–1763

2018
[20]

Xiao Lin, Xiaokai Chen, Chenyang Wang, Hantao Shu, Linfeng Song, Biao Li, and Peng Jiang. 2024. Discrete conditional diffusion for reranking in recommendation. InCompanion Proceedings of the ACM on Web Conference 2024. 161–169

2024
[21]

Zhijie Lin, Zhuofeng Li, Chenglei Dai, Wentian Bao, Shuai Lin, Enyun Yu, Haoxi- ang Zhang, and Liang Zhao. 2025. GReF: A Unified Generative Framework for Efficient Reranking via Ordered Multi-token Prediction. InProceedings of the 34th ACM International Conference on Information and Knowledge Management. 5879–5887

2025
[22]

Zihan Lin, Hui Wang, Jingshu Mao, Wayne Xin Zhao, Cheng Wang, Peng Jiang, and Ji-Rong Wen. 2022. Feature-aware diversified re-ranking with disentangled representations for relevant recommendation. InProceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining. 3327–3335

2022
[23]

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. 1524–1534

2023
[24]

Gurmeet Singh Manku, Arvind Jain, and Anish Das Sarma. 2007. Detecting near- duplicates for web crawling. InProceedings of the 16th international conference on World Wide Web. 141–150

2007
[25]

Liang Pang, Jun Xu, Qingyao Ai, Yanyan Lan, Xueqi Cheng, and Jirong Wen
[26]

InProceedings of the 43rd international ACM SIGIR conference on research and development in information retrieval

Setrank: Learning a permutation-invariant ranking model for information retrieval. InProceedings of the 43rd international ACM SIGIR conference on research and development in information retrieval. 499–508
[27]

Changhua Pei, Yi Zhang, Yongfeng Zhang, Fei Sun, Xiao Lin, Hanxiao Sun, Jian Wu, Peng Jiang, Junfeng Ge, Wenwu Ou, et al. 2019. Personalized re-ranking for recommendation. InProceedings of the 13th ACM conference on recommender systems. 3–11

2019
[28]

Qi Pi, Guorui Zhou, Yujing Zhang, Zhe Wang, Lejian Ren, Ying Fan, Xiaoqiang Zhu, and Kun Gai. 2020. Search-based user interest modeling with lifelong sequential behavior data for click-through rate prediction. InProceedings of the 29th ACM International Conference on Information & Knowledge Management. 2685–2692

2020
[29]

Shashank Rajput, Nikhil Mehta, Anima Singh, Raghunandan Hulikal Keshavan, Trung Vu, Lukasz Heldt, Lichan Hong, Yi Tay, Vinh Tran, Jonah Samost, et al
[30]

Information Processing Systems36 (2023), 10299–10315

Recommender systems with generative retrieval.Advances in Neural Conference’17, July 2017, Washington, DC, USA Trovato and Tobin, et al. Information Processing Systems36 (2023), 10299–10315

2017
[31]

Yuxin Ren, Qiya Yang, Yichun Wu, Wei Xu, Yalong Wang, and Zhiqiang Zhang
[32]

In Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining

Non-autoregressive generative models for reranking recommendation. In Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining. 5625–5634
[33]

Zhihong Shao, Peiyi Wang, Qihao Zhu, Runxin Xu, Junxiao Song, Xiao Bi, Haowei Zhang, Mingchuan Zhang, YK Li, Yang Wu, et al. 2024. Deepseekmath: Pushing the limits of mathematical reasoning in open language models.arXiv preprint arXiv:2402.03300(2024)

work page internal anchor Pith review Pith/arXiv arXiv 2024
[34]

Xiaowen Shi, Fan Yang, Ze Wang, Xiaoxu Wu, Muzhi Guan, Guogang Liao, Wang Yongkang, Xingxing Wang, and Dong Wang. 2023. PIER: Permutation-Level Interest-Based End-to-End Re-ranking Framework in E-commerce. InProceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining. 4823–4831

2023
[35]

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

2017
[36]

Shuli Wang, Yinqiu Huang, Changhao Li, Yuan Zhou, Yonggang Liu, Yongqiang Zhang, Yinhua Zhu, Haitao Wang, and Xingxing Wang. 2025. You Only Evaluate Once: A Tree-based Rerank Method at Meituan.arXiv preprint arXiv:2508.14420 (2025)

work page arXiv 2025
[37]

Shuli Wang, Xue Wei, Senjie Kou, Chi Wang, Wenshuai Chen, Qi Tang, Yinhua Zhu, Xiong Xiao, and Xingxing Wang. 2025. NLGR: Utilizing Neighbor Lists for Generative Rerank in Personalized Recommendation Systems. InCompanion Proceedings of the ACM on Web Conference 2025. 530–537

2025
[38]

Yunjia Xi, Weiwen Liu, Xinyi Dai, Ruiming Tang, Weinan Zhang, Qing Liu, Xi- uqiang He, and Yong Yu. 2021. Context-aware reranking with utility maximization for recommendation.arXiv preprint arXiv:2110.09059(2021)

work page arXiv 2021
[39]

Yunjia Xi, Weiwen Liu, Jieming Zhu, Xilong Zhao, Xinyi Dai, Ruiming Tang, Weinan Zhang, Rui Zhang, and Yong Yu. 2022. Multi-Level Interaction Reranking with User Behavior History. InProceedings of the 45th International ACM SIGIR Conference on Research and Development in Information Retrieval. 1336–1346

2022
[40]

Wencai Ye, Mingjie Sun, Shaoyun Shi, Peng Wang, Wenjin Wu, and Peng Jiang
[41]

DAS: Dual-Aligned Semantic IDs Empowered Industrial Recommender System.arXiv preprint arXiv:2508.10584(2025)

work page arXiv 2025
[42]

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

work page internal anchor Pith review arXiv 2024
[43]

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

2019
[44]

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. InProceedings of the 24th ACM SIGKDD international conference on knowledge discovery & data mining. 1059–1068

2018
[46]

Tao Zhuang, Wenwu Ou, and Zhirong Wang. 2018. Globally optimized mutual influence aware ranking in e-commerce search.arXiv preprint arXiv:1805.08524 (2018)

work page arXiv 2018