arxiv: 2602.23978 · v2 · submitted 2026-02-27 · 💻 cs.IR

Recognition: no theorem link

Towards Efficient and Generalizable Retrieval: Adaptive Semantic Quantization and Residual Knowledge Transfer

Huimu Wang , Xingzhi Yao , Yiming Qiu , Qinghong Zhang , Haotian Wang , Yufan Cui , Songlin Wang , Sulong Xu

show 1 more author

Mingming Li

Authors on Pith no claims yet

Pith reviewed 2026-05-15 18:56 UTC · model grok-4.3

classification 💻 cs.IR

keywords semantic quantizationgenerative retrievaladaptive quantizationcold-start retrievalresidual knowledge transferhead-tail imbalanceindustrial search

0 comments

The pith

SA²CRQ uses entropy-based variable code lengths and head-item manifold regularization to reduce collisions for popular items while improving generalization for cold-start ones.

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

The paper introduces the SA²CRQ framework to resolve a core tension in semantic ID generative retrieval: data-rich head items produce ID collisions that blur features, while data-sparse tail items fragment into isolated points that hurt generalization. SARQ dynamically assigns longer discriminative codes to head items and shorter generalizable codes to tail items based on path entropy. ACRQ then anchors tail-item learning to a frozen semantic manifold derived from head items. If correct, this yields better end-to-end retrieval accuracy in skewed industrial data, especially for rare queries.

Core claim

The SA²CRQ framework combines Sequential Adaptive Residual Quantization, which allocates code lengths according to item path entropy to give longer IDs to head items and shorter IDs to tail items, with Anchored Curriculum Residual Quantization, which regularizes tail-item representations using a frozen semantic manifold learned from head items, producing consistent gains over baselines on industrial and public datasets particularly for cold-start retrieval.

What carries the argument

Sequential Adaptive Residual Quantization (SARQ) for entropy-driven variable-length code allocation paired with Anchored Curriculum Residual Quantization (ACRQ) that transfers structure from a frozen head-item semantic manifold to regularize tail items.

If this is right

Head items receive longer IDs that reduce collisions and preserve distinct features.
Tail and cold-start items receive shorter IDs plus manifold regularization that improves generalization.
The method delivers measurable gains on large-scale industrial search systems and multiple public datasets.
Improvements concentrate in cold-start retrieval where data sparsity is most severe.

Where Pith is reading between the lines

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

Variable-length semantic IDs may require new index structures that support mixed code depths without extra lookup cost.
The same head-to-tail manifold transfer could apply to other power-law domains such as recommender systems or language-model tokenization.
Periodic refresh of the frozen manifold without full retraining might further lift tail performance.

Load-bearing premise

A frozen semantic manifold learned from head items can regularize tail-item representations without introducing systematic bias or preventing capture of genuinely novel semantics.

What would settle it

On a held-out cold-start test set, independently trained tail representations would achieve higher retrieval accuracy than those regularized by the head-item manifold.

Figures

Figures reproduced from arXiv: 2602.23978 by Haotian Wang, Huimu Wang, Mingming Li, Qinghong Zhang, Songlin Wang, Sulong Xu, Xingzhi Yao, Yiming Qiu, Yufan Cui.

**Figure 1.** Figure 1: The framework of SA2CRQ. Conversely, the massive volume of cold-start and tail items suffers from isolated, one-to-one mappings, limiting model generalization. To address the above issues, we propose the Anchored Curriculum with Sequential Adaptive Residual Quantization Framework (𝑆𝐴2𝐶𝑅𝑄) ( [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗

**Figure 3.** Figure 3: Performance under the sparse setting (w/o aug) [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗

**Figure 4.** Figure 4: The percent quantile of item count. RQ represents RQ-Kmeans. RQ TIGER ACRQ SARQ( =5e-7) SARQ( =1e-6) SARQ( =2e-6) SA²CRQ( =2e-6) 0 1 2 3 Sid Count 1e6 [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗

**Figure 7.** Figure 7: Comparison of distribution on head items. [PITH_FULL_IMAGE:figures/full_fig_p004_7.png] view at source ↗

read the original abstract

While semantic ID-based generative retrieval enables efficient end-to-end modeling in industrial applications, these methods face a persistent trade-off. On one hand, data-rich head items often suffer from ID collisions, which blur their distinct features and degrade downstream tasks. On the other hand, data-sparse tail items especially cold-start items are prone to semantic fragmentation during quantization; they are often mapped as isolated discrete points, which severely hinders their ability to generalize. To address this issue, we propose the Anchored Curriculum with Sequential Adaptive Quantization ($SA^2CRQ$) framework. The framework introduces Sequential Adaptive Residual Quantization (SARQ) to dynamically allocate code lengths based on item path entropy, assigning longer, discriminative IDs to head items and shorter, generalizable IDs to tail items. To mitigate data sparsity, the Anchored Curriculum Residual Quantization (ACRQ) component utilizes a frozen semantic manifold learned from head items to regularize and accelerate the representation learning of tail items. Experimental results from a large-scale industrial search system and multiple public datasets indicate that $SA^2CRQ$ yields consistent improvements over existing baselines, particularly in cold-start retrieval scenarios.

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

The paper combines entropy-based adaptive code allocation with head-manifold anchoring to handle head-tail imbalance in semantic ID quantization, but the abstract gives no numbers or controls to judge whether the claimed cold-start gains are real.

read the letter

The main takeaway is that SA²CRQ tries to solve the head-tail trade-off in generative retrieval by using path entropy to give longer, more discriminative IDs to popular items and shorter ones to rare items, then regularizing the tail representations with a frozen semantic manifold learned from the head. The specific pairing of sequential adaptive residual quantization and anchored curriculum regularization is the concrete new element here, and it is motivated directly from observable data distributions rather than from the evaluation metric itself. That framing is reasonable and targets a genuine production pain point where head collisions blur distinctions and tail items fragment into isolated points that hurt generalization. Reporting results on both public datasets and a large-scale industrial search system is the right move for relevance. The approach looks like honest engineering aimed at cold-start scenarios. The soft spots are straightforward. The abstract claims consistent improvements but supplies no effect sizes, no statistical tests, and no ablation details, so it is impossible to tell how much the method actually moves the needle or whether the baselines were competitive. The load-bearing assumption—that a frozen head manifold can regularize tail items without introducing bias or suppressing novel semantics—remains untested in the provided information. If tail distributions sit outside the head manifold, the anchoring could simply reduce fragmentation by pulling representations toward head modes rather than enabling genuine generalization. The stress-test note on this point is fair and needs direct checks such as manifold coverage or head-tail divergence metrics in the full experiments. This work is aimed at researchers in information retrieval who build generative models for search and recommendation. Someone already working on quantization or long-tail handling would get practical ideas from the mechanisms, though they would need the full experimental section before treating the gains as settled. It deserves peer review so the claims can be examined with proper controls and ablations.

Referee Report

2 major / 2 minor

Summary. The manuscript proposes the SA²CRQ framework for semantic ID-based generative retrieval. SARQ dynamically allocates code lengths via item path entropy to assign longer discriminative IDs to head items and shorter generalizable IDs to tail items. ACRQ employs a frozen semantic manifold learned from head items to regularize and accelerate representation learning for data-sparse tail items, including cold-start cases. Experiments on a large-scale industrial search system and multiple public datasets report consistent improvements over baselines, with particular gains in cold-start retrieval.

Significance. If the results hold under rigorous controls, the framework offers a practical advance in balancing ID collisions for head items against fragmentation for tail items in generative retrieval. The entropy-driven allocation and anchored regularization provide a data-driven mechanism for long-tail handling that could translate to measurable efficiency gains in industrial systems.

major comments (2)

[ACRQ description and experimental results] The central generalization claim for cold-start items rests on ACRQ's frozen head-item manifold transferring without systematic bias. The manuscript must supply explicit checks (e.g., embedding divergence statistics or manifold coverage metrics between head and tail distributions) in the experimental section; absent these, observed gains could arise from reduced fragmentation rather than true semantic transfer.
[Experimental results] The abstract states 'consistent improvements' without reporting effect sizes, statistical significance, or ablation controls. The results section must include these quantities (with confidence intervals and baseline comparisons) to establish that gains survive proper error analysis and are not artifacts of the chosen metrics.

minor comments (2)

[SARQ component] Define 'item path entropy' formally with an equation in the SARQ section; the current description is informal and prevents exact reproduction.
[ACRQ component] Clarify the precise procedure for learning and freezing the semantic manifold in ACRQ, including any hyperparameters and the stopping criterion for curriculum stages.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments, which help strengthen the presentation of our SA²CRQ framework. We address each major comment below and will incorporate the requested analyses in the revised manuscript.

read point-by-point responses

Referee: [ACRQ description and experimental results] The central generalization claim for cold-start items rests on ACRQ's frozen head-item manifold transferring without systematic bias. The manuscript must supply explicit checks (e.g., embedding divergence statistics or manifold coverage metrics between head and tail distributions) in the experimental section; absent these, observed gains could arise from reduced fragmentation rather than true semantic transfer.

Authors: We agree that explicit checks are required to isolate the contribution of semantic transfer from reduced fragmentation. In the revised experimental section we will report (i) embedding divergence statistics including mean cosine similarity and KL divergence between head-item and tail-item distributions in the frozen manifold, and (ii) manifold coverage metrics such as the fraction of tail embeddings lying inside the convex hull of head embeddings. These additions will directly address the concern and confirm that gains for cold-start items arise from anchored regularization rather than quantization effects alone. revision: yes
Referee: [Experimental results] The abstract states 'consistent improvements' without reporting effect sizes, statistical significance, or ablation controls. The results section must include these quantities (with confidence intervals and baseline comparisons) to establish that gains survive proper error analysis and are not artifacts of the chosen metrics.

Authors: We will expand the results section to include effect sizes (relative percentage improvements), statistical significance via paired t-tests or Wilcoxon signed-rank tests with reported p-values, 95% confidence intervals obtained by bootstrapping, and comprehensive ablation controls that isolate SARQ and ACRQ contributions against all baselines. These quantities will be presented for both the industrial dataset and public benchmarks to demonstrate that improvements are robust and not metric-specific artifacts. revision: yes

Circularity Check

0 steps flagged

No circularity: SA²CRQ components defined from observable data distributions

full rationale

The framework defines SARQ code allocation via item path entropy computed directly from data frequencies and ACRQ regularization via a frozen manifold extracted from head-item embeddings. Neither step reduces to the target retrieval metric by construction, nor relies on self-citation chains or fitted parameters renamed as predictions. Experimental gains are reported against external baselines on industrial and public datasets, keeping the derivation self-contained.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 2 invented entities

The framework rests on the assumption that item path entropy is a reliable proxy for the optimal code length and that a semantic manifold learned from head items transfers useful structure to tail items. No free parameters are explicitly named in the abstract, but the entropy threshold and curriculum schedule are likely fitted. No new physical entities are introduced.

axioms (2)

domain assumption A semantic manifold learned from head items provides useful regularization for tail items without introducing harmful bias.
Invoked in the description of the ACRQ component.
domain assumption Item path entropy accurately indicates the degree of feature distinctiveness needed for discriminative IDs.
Used to decide code length allocation in SARQ.

invented entities (2)

Sequential Adaptive Residual Quantization (SARQ) no independent evidence
purpose: Dynamically allocate code lengths based on item path entropy
New component introduced to handle head-tail imbalance in ID assignment.
Anchored Curriculum Residual Quantization (ACRQ) no independent evidence
purpose: Use frozen head-item manifold to regularize tail-item learning
New component to mitigate data sparsity for cold-start items.

pith-pipeline@v0.9.0 · 5526 in / 1386 out tokens · 22249 ms · 2026-05-15T18:56:02.016972+00:00 · methodology

discussion (0)

Forward citations

Cited by 1 Pith paper

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

CapsID: Soft-Routed Variable-Length Semantic IDs for Generative Recommendation
cs.IR 2026-05 unverdicted novelty 6.0

CapsID uses probabilistic capsule routing and confidence-based termination to generate variable-length semantic IDs, improving recall by 9.6% over strong baselines with half the latency of dual-representation systems.

Reference graph

Works this paper leans on

34 extracted references · 34 canonical work pages · cited by 1 Pith paper · 3 internal anchors

[1]

Payal Bajaj, Daniel Campos, Nick Craswell, Li Deng, Jianfeng Gao, Xiaodong Liu, Rangan Majumder, Andrew McNamara, Bhaskar Mitra, Tri Nguyen, Mir Rosenberg, Xia Song, Alina Stoica, Saurabh Tiwary, and Tong Wang. 2018. MS MARCO: A Human Generated MAchine Reading COmprehension Dataset. arXiv:1611.09268 [cs.CL] https://arxiv.org/abs/1611.09268

work page internal anchor Pith review Pith/arXiv arXiv 2018
[2]

Michele Bevilacqua, Giuseppe Ottaviano, Patrick Lewis, Scott Yih, Sebastian Riedel, and Fabio Petroni. 2022. Autoregressive search engines: Generating substrings as document identifiers.Advances in Neural Information Processing Systems35 (2022), 31668–31683

work page 2022
[3]

Ben Chen, Xian Guo, Siyuan Wang, Zihan Liang, Yue Lv, Yufei Ma, Xinlong Xiao, Bowen Xue, Xuxin Zhang, Ying Yang, et al . 2025. Onesearch: A preliminary exploration of the unified end-to-end generative framework for e-commerce search.arXiv preprint arXiv:2509.03236(2025)

work page arXiv 2025
[4]

Jiehan Cheng, Zhicheng Dou, Yutao Zhu, and Xiaoxi Li. 2025. Descriptive and Discriminative Document Identifiers for Generative Retrieval. InProceedings of the AAAI Conference on Artificial Intelligence, Vol. 39. 11518–11526

work page 2025
[6]

Jiaxin Deng, Shiyao Wang, Kuo Cai, Lejian Ren, Qigen Hu, Weifeng Ding, Qiang Luo, and Guorui Zhou. 2025. Onerec: Unifying retrieve and rank with generative recommender and iterative preference alignment.arXiv preprint arXiv:2502.18965 (2025)

work page internal anchor Pith review Pith/arXiv arXiv 2025
[7]

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. InKDD

work page 2025
[8]

Vladimir Karpukhin, Barlas Oguz, Sewon Min, Patrick SH Lewis, Ledell Wu, Sergey Edunov, Danqi Chen, and Wen-tau Yih. 2020. Dense Passage Retrieval for Open-Domain Question Answering.. InEMNLP (1). 6769–6781

work page 2020
[9]

Tongyoung Kim, Soojin Yoon, Seongku Kang, Jinyoung Yeo, and Dongha Lee

work page
[10]

MVIGER: Multi-View Variational Integration of Complementary Knowledge for Generative Recommender

SC-Rec: Enhancing Generative Retrieval with Self-Consistent Reranking for Sequential Recommendation. arXiv:2408.08686 [cs.IR] https://arxiv.org/abs/ 2408.08686

work page internal anchor Pith review Pith/arXiv arXiv
[11]

Mingming Li, Huimu Wang, Zuxu Chen, Guangtao Nie, Yiming Qiu, Guoyu Tang, Lin Liu, and Jingwei Zhuo. 2024. Generative Retrieval with Preference Optimization for E-commerce Search. arXiv:2407.19829 [cs.IR] https://arxiv.org/ abs/2407.19829

work page arXiv 2024
[12]

Mingming Li, Chunyuan Yuan, Huimu Wang, Peng Wang, Jingwei Zhuo, Binbin Wang, Lin Liu, and Sulong Xu. 2023. Adaptive Hyper-parameter Learning for Deep Semantic Retrieval. InProceedings of the 2023 Conference on Empirical Methods in Natural Language Processing: Industry Track. 775–782

work page 2023
[13]

Yongqi Li, Nan Yang, Liang Wang, Furu Wei, and Wenjie Li. 2023. Generative retrieval for conversational question answering.Information Processing & Man- agement60, 5 (2023), 103475

work page 2023
[14]

Yongqi Li, Nan Yang, Liang Wang, Furu Wei, and Wenjie Li. 2023. Learning to rank in generative retrieval.arXiv preprint arXiv:2306.15222(2023)

work page arXiv 2023
[16]

Yongqi Li, Nan Yang, Liang Wang, Furu Wei, and Wenjie Li. 2023. Multiview identifiers enhanced generative retrieval.arXiv preprint arXiv:2305.16675(2023). SIGIR ’26, July 20–24, 2026, Melbourne, VIC, Australia Wang Huimu et al

work page arXiv 2023
[17]

Han Liu, Yinwei Wei, Xuemeng Song, Weili Guan, Yuan-Fang Li, and Liqiang Nie. 2024. MMGRec: Multimodal Generative Recommendation with Transformer Model. arXiv:2404.16555 [cs.IR] https://arxiv.org/abs/2404.16555

work page arXiv 2024
[18]

Ilya Loshchilov and Frank Hutter. [n. d.]. Decoupled Weight Decay Regularization. InInternational Conference on Learning Representations

work page
[19]

Jianmo Ni, Gustavo Hernandez Abrego, Noah Constant, Ji Ma, Keith Hall, Daniel Cer, and Yinfei Yang. 2022. Sentence-T5: Scalable Sentence Encoders from Pre- trained Text-to-Text Models. InFindings of the Association for Computational Linguistics: ACL 2022, Smaranda Muresan, Preslav Nakov, and Aline Villavicencio (Eds.). Association for Computational Lingui...

work page doi:10.18653/v1/2022.findings-acl.146 2022
[20]

Yiming Qiu, Kang Zhang, Han Zhang, Songlin Wang, Sulong Xu, Yun Xiao, Bo Long, and Wen-Yun Yang. 2021. Query rewriting via cycle-consistent transla- tion for e-commerce search. In2021 IEEE 37th International Conference on Data Engineering (ICDE). IEEE, 2435–2446

work page 2021
[21]

Tran, Jonah Samost, Maciej Kula, Ed H

Shashank Rajput, Nikhil Mehta, Anima Singh, Raghunandan Hulikal Keshavan, Trung Vu, Lukasz Heldt, Lichan Hong, Yi Tay, Vinh Q. Tran, Jonah Samost, Maciej Kula, Ed H. Chi, and Maheswaran Sathiamoorthy. 2023. Recommender Systems with Generative Retrieval. InThirty-seventh Conference on Neural Information Processing Systems. https://openreview.net/forum?id=B...

work page 2023
[22]

Ruiyang Ren, Wayne Xin Zhao, Jing Liu, Hua Wu, Ji-Rong Wen, and Haifeng Wang. 2023. TOME: A Two-stage Approach for Model-based Retrieval.arXiv preprint arXiv:2305.11161(2023)

work page arXiv 2023
[23]

Stephen Robertson, Hugo Zaragoza, et al . 2009. The probabilistic relevance framework: BM25 and beyond.Foundations and Trends®in Information Retrieval 3, 4 (2009), 333–389

work page 2009
[25]

Weiwei Sun, Lingyong Yan, Zheng Chen, Shuaiqiang Wang, Haichao Zhu, Pengjie Ren, Zhumin Chen, Dawei Yin, Maarten Rijke, and Zhaochun Ren. 2024. Learning to tokenize for generative retrieval.Advances in Neural Information Processing Systems36 (2024)

work page 2024
[26]

Yubao Tang, Ruqing Zhang, Jiafeng Guo, Jiangui Chen, Zuowei Zhu, Shuaiqiang Wang, Dawei Yin, and Xueqi Cheng. 2023. Semantic-Enhanced Differentiable Search Index Inspired by Learning Strategies.arXiv preprint arXiv:2305.15115 (2023)

work page arXiv 2023
[27]

Yi Tay, Vinh Tran, Mostafa Dehghani, Jianmo Ni, Dara Bahri, Harsh Mehta, Zhen Qin, Kai Hui, Zhe Zhao, Jai Gupta, et al. 2022. Transformer memory as a differentiable search index.Advances in Neural Information Processing Systems 35 (2022), 21831–21843

work page 2022
[28]

Wenjie Wang, Honghui Bao, Xinyu Lin, Jizhi Zhang, Yongqi Li, Fuli Feng, See- Kiong Ng, and Tat-Seng Chua. 2024. Learnable Item Tokenization for Generative Recommendation. InInternational Conference on Information and Knowledge Management

work page 2024
[29]

Yujing Wang, Yingyan Hou, Haonan Wang, Ziming Miao, Shibin Wu, Qi Chen, Yuqing Xia, Chengmin Chi, Guoshuai Zhao, Zheng Liu, et al . 2022. A neural corpus indexer for document retrieval.Advances in Neural Information Processing Systems35 (2022), 25600–25614

work page 2022
[30]

Ye Wang, Jiahao Xun, Minjie Hong, Jieming Zhu, Tao Jin, Wang Lin, Haoyuan Li, Linjun Li, Yan Xia, Zhou Zhao, and Zhenhua Dong. 2024. EAGER: Two-Stream Generative Recommender with Behavior-Semantic Collaboration. InProceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (Barcelona, Spain)(KDD ’24). Association for Computing Mac...

work page doi:10.1145/3637528.3671775 2024
[31]

Zihan Wang, Yujia Zhou, Yiteng Tu, and Zhicheng Dou. 2023. NOVO: Learnable and Interpretable Document Identifiers for Model-Based IR. InProceedings of the 32nd ACM International Conference on Information and Knowledge Manage- ment(Birmingham, United Kingdom)(CIKM ’23). Association for Computing Machinery, New York, NY, USA, 2656–2665. doi:10.1145/3583780.3614993

work page doi:10.1145/3583780.3614993 2023
[32]

Peiwen Yuan, Xinglin Wang, Shaoxiong Feng, Boyuan Pan, Yiwei Li, Heda Wang, Xupeng Miao, and Kan Li. 2024. Generative Dense Retrieval: Memory Can Be a Burden.arXiv preprint arXiv:2401.10487(2024)

work page arXiv 2024
[33]

Peitian Zhang, Zheng Liu, Yujia Zhou, Zhicheng Dou, Fangchao Liu, and Zhao Cao. 2024. Generative Retrieval via Term Set Generation. InProceedings of the 47th International ACM SIGIR Conference on Research and Development in Infor- mation Retrieval(Washington DC, USA)(SIGIR ’24). Association for Computing Machinery, New York, NY, USA, 458–468. doi:10.1145/...

work page doi:10.1145/3626772.3657797 2024
[34]

Bowen Zheng, Yupeng Hou, Hongyu Lu, Yu Chen, Wayne Xin Zhao, Ming Chen, and Ji-Rong Wen. 2024. Adapting Large Language Models by Integrating Collaborative Semantics for Recommendation. In2024 IEEE 40th International Conference on Data Engineering (ICDE). 1435–1448. doi:10.1109/ICDE60146.2024. 00118

work page doi:10.1109/icde60146.2024 2024
[35]

Carolina Zheng, Minhui Huang, Dmitrii Pedchenko, Kaushik Rangadurai, Siyu Wang, Gaby Nahum, Jie Lei, Yang Yang, Tao Liu, Zutian Luo, Xiaohan Wei, Dinesh Ramasamy, Jiyan Yang, Yiping Han, Lin Yang, Hangjun Xu, Rong Jin, and Shuang Yang. 2025. Enhancing Embedding Representation Stability in Recommendation Systems with Semantic ID. arXiv:2504.02137 [cs.IR] h...

work page arXiv 2025
[36]

Yujia Zhou, Jing Yao, Zhicheng Dou, Ledell Wu, Peitian Zhang, and Ji-Rong Wen

work page
[37]

Company Portrait JD.com, Inc., also known as Jingdong, is a Chinese e-commerce company headquartered in Beijing

Ultron: An ultimate retriever on corpus with a model-based indexer.arXiv preprint arXiv:2208.09257(2022). Company Portrait JD.com, Inc., also known as Jingdong, is a Chinese e-commerce company headquartered in Beijing. It is one of the two massive B2C online retailers in China by transaction volume and revenue, a member of the Fortune Global 500. When cla...

work page arXiv 2022