Pith. sign in

REVIEW 4 major objections 4 minor 2 cited by

This paper releases KuaiSearch, a dataset of over two million real e-commerce search queries with plaintext product descriptions, built to support recall, ranking, and relevance research.

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 ↗

T0 review · deepseek-v4-flash

2026-08-03 00:06 UTC pith:V4A6IPY7

load-bearing objection A genuinely useful e-commerce search resource whose 'largest' claim needs to be checked against Amazon ESCI and whose sampling details need tightening. the 4 major comments →

arxiv 2602.11518 v3 pith:V4A6IPY7 submitted 2026-02-12 cs.IR

KuaiSearch: An E-Commerce Search Dataset with Authentic Queries and Product Texts for Recall, Ranking, and Relevance

Yupeng Li , Ben Chen , Mingyue Cheng , Zhiding Liu , Xuxin Zhang , Chenyi Lei This is my paper
classification cs.IR
keywords E-commerce searchDatasetRecallRankingRelevanceCold-start usersLong-tail productsLarge language models
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.

KuaiSearch is presented as the largest e-commerce search dataset currently available, built from real user search interactions on a major Chinese short-video platform. It keeps authentic user queries and plaintext product texts, and deliberately does not filter out cold-start users or long-tail products, so the data distribution mirrors a real commercial environment. The dataset provides separate benchmarks for the three main stages of the search pipeline—recall, ranking, and relevance judgment—so a single resource can support end-to-end research. Benchmarks on representative models show the expected pattern: embedding retrieval beats lexical and generative baselines on recall, attention-based models lead on ranking, and larger LLMs do best on relevance. The value of the paper is in releasing this resource to the research community.

Core claim

The central claim is that KuaiSearch is the largest publicly available e-commerce search dataset, with real user queries and natural-language product texts rather than heuristically constructed or anonymized content. It covers approximately 331,930 users, 18.6 million products, and 2.57 million queries, and it spans recall, ranking, and relevance tasks within a single consistent data source. By retaining cold-start users and long-tail products without popularity-based filtering, the authors argue the dataset faithfully reflects real-world e-commerce search distributions, making it suitable for training and evaluating LLM-based search models.

What carries the argument

The dataset itself is the central artifact. Its recall component records search requests with impressed, clicked, and purchased item ids; its ranking component adds user and item statistical features plus the last 20 clicked and purchased items; and its relevance component contains 46,422 query–product pairs annotated by experts on a four-level graded scale. The three benchmarks built on these components carry the experimental argument: recall methods are compared with R@K and HR@K on the last day's logs, ranking methods with Logloss and ROC-AUC for CTR prediction, and relevance methods with ROC-AUC and PR-AUC for binary classification.

Load-bearing premise

The dataset's realism rests on the assumption that the unspecified criteria for 'core interaction behaviors' used to sample the 331,930 users do not silently exclude the casual or inactive shoppers the paper claims to retain.

What would settle it

Inspect the sampling script or reproduce the sampling from the full logs; if the 'core interaction behaviors' filter excludes more than a trivial fraction of users with very few searches, the claim that no constraints were imposed on the user population fails.

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

If this is right

  • Researchers can train and evaluate LLM-based semantic matching on authentic, unfiltered queries and plaintext product titles, avoiding the artifacts of heuristically constructed query sets.
  • Cold-start users and long-tail products are part of the evaluation data, so models that rely on popularity signals will be tested under realistic sparsity.
  • Because the three pipeline stages share the same underlying user behavior logs, a single dataset can support end-to-end pipeline studies rather than separate, incompatible benchmarks.
  • The four-level graded relevance annotations provide a fine-grained supervision signal that goes beyond binary relevance, useful for attribute-level reasoning.

Where Pith is reading between the lines

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

  • The sampling step is only described as 'randomly sampled 331,930 users who exhibited core interaction behaviors after June 1, 2025' with an unspecified time window; if that filter is stricter than described, the 'no constraints' claim could overstate coverage of casual users. Researchers using the dataset should check the actual behavioral distribution.
  • The relevance benchmark is much smaller than the other two (46K pairs), so its primary use may be as a fine-tuning signal rather than a standalone evaluation set; combining it with weak supervision from clicks is a natural next step the paper does not explore.
  • With 99.82% of users located in China and queries in Chinese, the dataset's conclusions about model performance may not transfer directly to other languages or markets without additional validation.
  • The temporal split (last day as test) is a realistic but potentially optimistic setup; if user behavior drifts across days, models may look better than they would under a random split. A robustness check across different time windows would strengthen the benchmarks.

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 / 4 minor

Summary. The paper introduces KuaiSearch, a large e-commerce search dataset built from real user interaction logs on the Kuaishou platform. The dataset contains approximately 330,000 users, 18.6 million products, and 2.57 million real search queries, and is organized into three task components: recall, ranking, and relevance judgment. The authors claim that, to the best of their knowledge, this is the largest e-commerce search dataset currently available, and that it preserves authentic user queries and plaintext product texts without popularity-based filtering, thus covering cold-start users and long-tail products. The paper provides a multi-perspective analysis of the data (demographics, product/user/query distributions, relevance label distribution) and benchmarks representative retrieval, ranking, and relevance models on a lightweight subset, KuaiSearch-Lite.

Significance. If the dataset is released as described, it is a potentially valuable public resource for e-commerce search research, particularly for LLM-based approaches that require natural-language queries and product texts. The multi-stage coverage (recall, ranking, relevance) is a distinctive feature relative to prior single-stage datasets, and the inclusion of cold-start users and long-tail products addresses known limitations of older datasets such as Amazon. The paper ships a public GitHub link and provides descriptive statistics that are useful for the community. However, the headline 'largest dataset' claim is not currently supported by the presented comparisons, and some construction details are under-specified. The benchmark experiments are preliminary but cover a reasonable set of baselines.

major comments (4)
  1. [§1 (Abstract, Table 1, §6)] The central claim that KuaiSearch is 'the largest e-commerce search dataset currently available' is not substantiated. Table 1 compares only with Amazon and JDsearch, omitting Amazon ESCI / Shopping Queries (Reddy et al., 2022), which is a widely used public e-commerce search dataset with roughly 1.5M unique queries, 4.4M products, and ~43M query-product relevance pairs. The manuscript does not define the class of datasets under comparison nor the metric(s) on which 'largest' is judged (number of users, items, queries, interaction rows, or relevance pairs). Depending on the metric, ESCI may exceed KuaiSearch in some dimensions. The claim needs to be qualified, and the comparison table should enumerate the relevant datasets and specify the criteria.
  2. [§3, User Behavior Collection] The sampling protocol is under-specified. The text says 'randomly sampled 331,930 users who exhibited core interaction behaviors on the platform after June 1, 2025' and collected 'all product search activities within the corresponding time window,' but the length of the time window, the definition of 'core interaction behaviors,' and the fraction of eligible users excluded by this filter are not given. The later assertion that 'no constraints were imposed on the user population or the length of users' historical behaviors' appears to contradict the existence of this inclusion filter. Without these details, the claimed representativeness and the 'no popularity-based filtering' statement cannot be fully verified, and the dataset's coverage of casual or inactive users is unclear.
  3. [§4.3 and §5.3 (Relevance Data)] The relevance judgment component is a core contribution, but the annotation procedure and label quality are not sufficiently documented. The paper states that pairs were 'manually annotated by domain experts' with a four-level graded scheme, but gives no number of annotators, adjudication process, or inter-annotator agreement (e.g., Cohen's kappa or Krippendorff's alpha). Without such information, the reliability of the 46,422 relevance labels is unknown, and the relevance benchmark results in Table 8 are hard to interpret as ground truth for model comparison. Please provide annotation statistics and a brief protocol description.
  4. [§5 (Tables 6–8)] All benchmark results are reported as single runs without variance estimates or statistical significance tests. This is particularly problematic in Table 7, where the ROC-AUC differences between DIN and DNN are only 0.0004, and in Table 8, where Qwen3-0.6B and Llama3.2-3B are within 0.009 ROC-AUC. Without multiple seeds or significance testing, the qualitative conclusions (e.g., 'DIN performs the best in ROC-AUC', 'Qwen3-1.7B delivers the best overall performance') are not statistically supported. For a benchmark paper, reporting mean and standard deviation over several runs (or at least a statement that differences are not stable) is expected.
minor comments (4)
  1. [General formatting] The manuscript contains ACM template placeholders that should be cleaned before publication: the copyright line reads '©2018 Copyright held by the owner/author(s)', the conference metadata says 'Conference acronym ’XX, Woodstock, NY' and the reference list ends with 'Received 20 February 2007; revised 12 March 2009; accepted 5 June 2009.' These are presumably artifacts of the template but are misleading in the current version.
  2. [§1 / Availability] The dataset is said to be publicly available at a GitHub link, but no license is specified. A clear license (e.g., CC BY 4.0 or a custom research-only license) is important for reuse and should be stated in the paper.
  3. [§5.1 and §5.2] The train/test split is described only as using 'the last day' of the KuaiSearch-Lite data. This is ambiguous: does the last day refer to the last calendar day in the collection window, and what is the associated date range? Also, the size of the test set is not reported. Please clarify the split and report test-set statistics to enable reproducibility.
  4. [§4.2.3] The query category assignment is based on an aggregation heuristic (majority exposure category). This is a reasonable approximation, but the paper should acknowledge that this heuristic may mislabel ambiguous queries; for example, a query with mixed exposure across categories is assigned a single label. A brief caveat would be appropriate.

Circularity Check

0 steps flagged

No circularity: KuaiSearch is a dataset/benchmark paper whose central claims rest on externally logged interactions and human annotation, not on a derivation or fitted prediction.

full rationale

This is a dataset and benchmark paper, not a derivation. The central claims — that KuaiSearch is built from real Kuaishou user search logs, contains plaintext product texts and authentic queries, covers cold-start users and long-tail products, and spans recall, ranking, and relevance — are supported by external data collection and manual annotation, not by fitting a model to the data and then predicting the same data. The benchmark experiments (Sections 5.1–5.3) evaluate existing methods on held-out portions of the dataset and do not feed results back into dataset construction. The paper's self-citations ([4], [14], [21], [22], [40]) appear only in related-work discussion of generative search frameworks and are not load-bearing for the dataset's validity or scale. One notable weakness is correctness-related rather than circular: the 'largest e-commerce search dataset' claim (Abstract, Section 1, Section 6, Table 1) is comparative but omits comparators such as Amazon ESCI / Shopping Queries, and no metric or comparison class is defined. That is an evidential gap, not a case of a prediction reducing by construction to its inputs. No circular step can be exhibited from the paper's text, so the appropriate finding is no significant circularity.

Axiom & Free-Parameter Ledger

0 free parameters · 3 axioms · 0 invented entities

The central claim rests on the dataset being a faithful sample of real interactions and on the relevance labels being trustworthy. No fitted parameters or invented theoretical entities are required; benchmark hyperparameters (e.g., codebook sizes, number of pseudo-queries) are standard design choices that do not feed back into the dataset claim.

axioms (3)
  • domain assumption Kuaishou's e-commerce search logs are representative of e-commerce search in general.
    The paper argues KuaiSearch reflects real-world scenarios based on logs from one platform and one sampling rule (Section 3); if Kuaishou's user base or query mix is atypical, the generality claim weakens.
  • domain assumption Human expert relevance annotations are reliable ground truth.
    Section 4.3 treats the four-level labels as supervision without reporting annotator agreement, qualifications beyond 'domain experts,' or quality-control steps.
  • domain assumption The comparison set in Table 1 is sufficient to support the 'largest' claim.
    Only Amazon, JDsearch, and Qilin are compared; the Amazon ESCI/Shopping Queries dataset is not mentioned. Other non-public industrial logs are also excluded, but those are not citable benchmarks.

pith-pipeline@v1.3.0-alltime-deepseek · 13579 in / 14066 out tokens · 148350 ms · 2026-08-03T00:06:28.531291+00:00 · methodology

0 comments
read the original abstract

E-commerce search serves as a central interface connecting user demands with massive product inventories and plays a vital role in daily online shopping. However, it faces challenges, including highly ambiguous queries, noisy product texts with weak semantic order, and diverse user preferences, making it difficult to accurately capture user intent and fine-grained product semantics. Recent advances in large language models for semantic representation and contextual reasoning have created new opportunities to address these challenges. Nevertheless, existing e-commerce search datasets still suffer from notable limitations: queries are often heuristically constructed, cold-start users and long-tail products are filtered out, query and product texts are anonymized, and most datasets cover only a single stage of the search pipeline. These limitations hinder realistic and comprehensive evaluation and constrain research on LLM-based e-commerce search. To address them, we construct and release KuaiSearch, a large-scale e-commerce search dataset built upon real user interactions from the Kuaishou platform. KuaiSearch preserves authentic user queries and natural-language product texts, covers cold-start users and long-tail products, and provides dedicated benchmarks for three key tasks in the e-commerce search pipeline: recall, ranking, and relevance judgment. We conduct a comprehensive analysis of KuaiSearch from multiple perspectives, including products, users, and queries, and establish benchmarks across representative search tasks. Experimental results demonstrate that KuaiSearch provides a valuable foundation for real-world e-commerce search research. The dataset is publicly available at: https://github.com/benchen4395/KuaiSearch

Figures

Figures reproduced from arXiv: 2602.11518 by Ben Chen, Chenyi Lei, Mingyue Cheng, Xuxin Zhang, Yupeng Li, Zhiding Liu.

Figure 1
Figure 1. Figure 1: The main search entries in the Kuaishou Platform. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Log-log distributions of product interaction fre [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Word cloud visualizations of the most frequent tokens in item titles and user queries (Top-200), reflecting the lexical [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Distribution of the top-12 query categories in our dataset. The inner ring displays query categories and their proportions [PITH_FULL_IMAGE:figures/full_fig_p006_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 2 Pith papers

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

  1. RecRM-Bench: Benchmarking Multidimensional Reward Modeling for Agentic Recommender Systems

    cs.IR 2026-05 unverdicted novelty 7.0

    RecRM-Bench is a new large-scale benchmark dataset and framework for multi-dimensional reward modeling in agentic recommender systems, spanning instruction following, factual consistency, query-item relevance, and use...

  2. Valley3: Scaling Omni Foundation Models for E-commerce

    cs.AI 2026-05 unverdicted novelty 4.0

    Valley3 is an omni MLLM for e-commerce that uses a four-stage pre-training pipeline plus post-training for controllable reasoning and agentic search, outperforming baselines on e-commerce benchmarks while staying comp...

Reference graph

Works this paper leans on

45 extracted references · 7 linked inside Pith · cited by 2 Pith papers

  1. [1]

    Qingyao Ai, Daniel N Hill, SVN Vishwanathan, and W Bruce Croft. 2019. A zero attention model for personalized product search. InProceedings of the 28th ACM International Conference on Information and Knowledge Management. 379–388

    2019

  2. [2]

    Qingyao Ai and Lakshmi Narayanan. R. 2021. Model-agnostic vs. model-intrinsic interpretability for explainable product search. InProceedings of the 30th ACM International Conference on Information & Knowledge Management. 5–15

    2021

  3. [3]

    Qingyao Ai, Yongfeng Zhang, Keping Bi, Xu Chen, and W Bruce Croft. 2017. Learning a hierarchical embedding model for personalized product search. In Proceedings of the 40th International ACM SIGIR Conference on Research and Development in Information Retrieval. 645–654

    2017

  4. [4]

    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)

    arXiv 2025

  5. [5]

    Jia Chen, Qian Dong, Haitao Li, Xiaohui He, Yan Gao, Shaosheng Cao, Yi Wu, Ping Yang, Chen Xu, Yao Hu, et al. 2025. Qilin: A multimodal information retrieval dataset with app-level user sessions. InProceedings of the 48th International ACM SIGIR Conference on Research and Development in Information Retrieval. 3670–3680

    2025

  6. [6]

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

  7. [7]

    Alexis Conneau, Kartikay Khandelwal, Naman Goyal, Vishrav Chaudhary, Guil- laume Wenzek, Francisco Guzmán, Edouard Grave, Myle Ott, Luke Zettlemoyer, and Veselin Stoyanov. 2020. Unsupervised cross-lingual representation learning at scale. InProceedings of the 58th annual meeting of the association for computa- tional linguistics. 8440–8451

    2020

  8. [8]

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

    2016

  9. [9]

    Jose GC De Souza, Michael Kozielski, Prashant Mathur, Ernie Chang, Marco Guerini, Matteo Negri, Marco Turchi, and Evgeny Matusov. 2018. Generating e-commerce product titles and predicting their quality. InProceedings of the 11th international conference on natural language generation. 233–243

    2018

  10. [10]

    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)

    Pith/arXiv arXiv 2025

  11. [11]

    Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. Bert: Pre-training of deep bidirectional transformers for language understanding. In Proceedings of the 2019 conference of the North American chapter of the association for computational linguistics: human language technologies, volume 1 (long and short papers). 4171–4186

    2019

  12. [12]

    Zhe Dong, Jianmo Ni, Daniel M Bikel, Enrique Alfonseca, Yuan Wang, Chen Qu, and Imed Zitouni. 2022. Exploring dual encoder architectures for question answering.arXiv preprint arXiv:2204.07120(2022)

    Pith/arXiv arXiv 2022

  13. [13]

    Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Amy Yang, Angela Fan, et al. 2024. The llama 3 herd of models.arXiv e-prints(2024), arXiv–2407

    2024

  14. [14]

    Xian Guo, Ben Chen, Siyuan Wang, Ying Yang, Chenyi Lei, Yuqing Ding, and Han Li. 2025. OneSug: The Unified End-to-End Generative Framework for E-commerce Query Suggestion.arXiv preprint arXiv:2506.06913(2025)

    Pith/arXiv arXiv 2025

  15. [15]

    Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, Weizhu Chen, et al. 2022. Lora: Low-rank adaptation of large language models.ICLR1, 2 (2022), 3

    2022

  16. [16]

    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

    2020

  17. [17]

    Tom Kwiatkowski, Jennimaria Palomaki, Olivia Redfield, Michael Collins, Ankur Parikh, Chris Alberti, Danielle Epstein, Illia Polosukhin, Jacob Devlin, Kenton Lee, et al. 2019. Natural questions: a benchmark for question answering research. Transactions of the Association for Computational Linguistics7 (2019), 453–466

    2019

  18. [18]

    Yongqi Li, Nan Yang, Liang Wang, Furu Wei, and Wenjie Li. 2024. Learning to rank in generative retrieval. InProceedings of the AAAI Conference on Artificial Intelligence, Vol. 38. 8716–8723

    2024

  19. [19]

    Jiongnan Liu, Zhicheng Dou, Guoyu Tang, and Sulong Xu. 2023. Jdsearch: A personalized product search dataset with real queries and full interactions. In Proceedings of the 46th International ACM SIGIR Conference on Research and Development in Information Retrieval. 2945–2952

    2023

  20. [20]

    Jiongnan Liu, Zhicheng Dou, Qiannan Zhu, and Ji-Rong Wen. 2022. A category- aware multi-interest model for personalized product search. InProceedings of the acm web conference 2022. 360–368

    2022

  21. [21]

    Zhiding Liu, Ben Chen, Mingyue Cheng, Enhong Chen, Li Li, Chenyi Lei, Wenwu Ou, Han Li, and Kun Gai. 2025. Towards Context-aware Reasoning-enhanced Generative Searching in E-commerce.arXiv preprint arXiv:2510.16925(2025)

    arXiv 2025

  22. [22]

    Xinchen Luo, Jiangxia Cao, Tianyu Sun, Jinkai Yu, Rui Huang, Wei Yuan, Hezheng Lin, Yichen Zheng, Shiyao Wang, Qigen Hu, et al . 2025. Qarm: Quantitative alignment multi-modal recommendation at kuaishou. InProceedings of the 34th ACM International Conference on Information and Knowledge Management. 5915– 5922

    2025

  23. [23]

    Tri Nguyen, Mir Rosenberg, Xia Song, Jianfeng Gao, Saurabh Tiwary, Rangan Majumder, and Li Deng. 2016. Ms marco: A human-generated machine reading comprehension dataset. (2016)

    2016

  24. [24]

    Rodrigo Nogueira, Wei Yang, Jimmy Lin, and Kyunghyun Cho. 2019. Document expansion by query prediction.arXiv preprint arXiv:1904.08375(2019)

    Pith/arXiv arXiv 2019

  25. [25]

    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

    2009

  26. [26]

    Gerard Salton and Christopher Buckley. 1988. Term-weighting approaches in automatic text retrieval.Information processing & management24, 5 (1988), 513–523

    1988

  27. [27]

    Craig Silverstein, Hannes Marais, Monika Henzinger, and Michael Moricz. 1999. Analysis of a very large web search engine query log. InAcm sigir forum, Vol. 33. ACM New York, NY, USA, 6–12

    1999

  28. [28]

    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

    2022

  29. [29]

    Manos Tsagkias, Tracy Holloway King, Surya Kallumadi, Vanessa Murdock, and Maarten De Rijke. 2021. Challenges and research opportunities in ecommerce search and recommendations. InACM Sigir Forum, Vol. 54. ACM New York, NY, USA, 1–23

    2021

  30. [30]

    Christophe Van Gysel, Maarten de Rijke, and Evangelos Kanoulas. 2016. Learning latent vector spaces for product search. InProceedings of the 25th ACM interna- tional on conference on information and knowledge management. 165–174

    2016

  31. [31]

    Jingang Wang, Junfeng Tian, Long Qiu, Sheng Li, Jun Lang, Luo Si, and Man Lan

  32. [32]

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

    2017

  33. [33]

    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. 1785–1797

    2021

  34. [34]

    Shitao Xiao, Zheng Liu, Peitian Zhang, Niklas Muennighoff, Defu Lian, and Jian-Yun Nie. 2024. C-pack: Packed resources for general chinese embeddings. InProceedings of the 47th international ACM SIGIR conference on research and development in information retrieval. 641–649

    2024

  35. [35]

    Lee Xiong, Chenyan Xiong, Ye Li, Kwok-Fung Tang, Jialin Liu, Paul Bennett, Junaid Ahmed, and Arnold Overwijk. 2020. Approximate nearest neighbor nega- tive contrastive learning for dense text retrieval.arXiv preprint arXiv:2007.00808 (2020)

    Pith/arXiv arXiv 2020

  36. [36]

    Linting Xue, Noah Constant, Adam Roberts, Mihir Kale, Rami Al-Rfou, Aditya Siddhant, Aditya Barua, and Colin Raffel. 2021. mT5: A massively multilingual pre-trained text-to-text transformer. InProceedings of the 2021 conference of the North American chapter of the association for computational linguistics: Human language technologies. 483–498

    2021

  37. [37]

    An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, et al. 2025. Qwen3 technical report.arXiv preprint arXiv:2505.09388(2025)

    Pith/arXiv arXiv 2025

  38. [38]

    Yuki Yano, Yukihiro Tagami, and Akira Tajima. 2016. Quantifying query ambi- guity with topic distributions. InProceedings of the 25th ACM International on Conference on Information and Knowledge Management. 1877–1880

    2016

  39. [39]

    Xue Zhang, Fusen Guo, Tao Chen, Lei Pan, Gleb Beliakov, and Jianzhang Wu

  40. [40]

    Zexin Zheng, Huangyu Dai, Lingtao Mao, Xinyu Sun, Zihan Liang, Ben Chen, Yuqing Ding, Chenyi Lei, Wenwu Ou, Han Li, et al. 2025. OneVision: An End-to- End Generative Framework for Multi-view E-commerce Vision Search.arXiv preprint arXiv:2510.05759(2025)

    arXiv 2025

  41. [41]

    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

  42. [42]

    Shengyao Zhuang, Houxing Ren, Linjun Shou, Jian Pei, Ming Gong, Guido Zuc- con, and Daxin Jiang. 2022. Bridging the gap between indexing and retrieval for differentiable search index with query generation.arXiv preprint arXiv:2206.10128 (2022). Received 20 February 2007; revised 12 March 2009; accepted 5 June 2009

    Pith/arXiv arXiv 2022

  43. [2016]

    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

  44. [2018]

    InProceedings of the AAAI conference on artificial intelligence, Vol

    A multi-task learning approach for improving product title compression with user search log data. InProceedings of the AAAI conference on artificial intelligence, Vol. 32

  45. [2023]

    A brief survey of machine learning and deep learning techniques for e-commerce research.Journal of Theoretical and Applied Electronic Commerce Research18, 4 (2023), 2188–2216

    2023