Pith. sign in

REVIEW 4 major objections 5 minor 67 references

This paper claims one generative vision-language model can run e-commerce search end-to-end, with up to +13.61% online GMV lift.

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

2026-08-01 17:47 UTC pith:4MYWYXK4

load-bearing objection HybSID is a real contribution to semantic-ID retrieval; the +13.61% GMV headline is an unsupported sum of two different floor-level A/B deployments. the 4 major comments →

arxiv 2607.17499 v1 pith:4MYWYXK4 submitted 2026-07-20 cs.AI

Pailitao-MMSearch: Building Native E-Commerce Multimodal Search Foundation

Xiaohan Ye , Xu Chen , Zihan Gong , Jian Ding , Lianyu Du , Baicheng Chen , Yunmeng Shu , Jingqian Zhao
show 10 more authors
Zhixiang Zhao Shuaiqi Jia Chong Ma Shuwen Xiao Xiangheng Kong Yuan Gao Jun Song Jinsong Lan Xiaoyong Zhu Bo Zheng
This is my paper
classification cs.AI
keywords multimodal searchgenerative retrievalsemantic IDse-commercevision-language modelcontinual pre-trainingreinforcement learningproduct tokenization
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.

Pailitao-MMSearch claims that e-commerce search, which today relies on separate text, image, and voice models patched together, can be handled by one generative vision-language model. The paper introduces HybSID, a product tokenization that joins three discrete semantic codes with a continuous multimodal embedding, so the model can both generalize across similar products and tell near-identical ones apart. A two-stage continual pre-training first infuses product and user-behavior knowledge, then uses on-policy distillation from the original base model to keep general language skills. A hybrid reasoning post-training pipeline adds direct generation, chain-of-thought for hard intents, and reinforcement learning with verifiable product rewards. Deployed on Taobao's Pailitao platform, it reports up to +13.61% GMV and +8.21% transaction volume, though the largest figure is a floor-specific sum and platform-wide lifts are near 1%.

Core claim

The paper's central claim: a single autoregressive vision-language model can serve as a complete e-commerce multimodal search engine, retrieving and ranking products end-to-end, if products are represented as HybSID — three residual-quantized discrete semantic codes plus one continuous multimodal embedding emitted through a shared <|emb_token|> token. Discrete codes let the model generate coarse-to-fine product groups efficiently; the continuous embedding recovers fine-grained distinctions quantization destroys, and nearest-neighbor search over the decoded embedding finishes ranking. The authors argue this removes the need to decompose image-plus-text queries into single-modal sub-queries. T

What carries the argument

HybSID (Hybrid Semantic ID) is the load-bearing object: each product becomes a token sequence of three residual-quantized discrete semantic codes (codebooks of 8192 entries each) plus one continuous multimodal embedding emitted at a special <|emb_token|> token. HybProjection maps an input product's continuous embedding into the model's representation space; HybDecoder maps the hidden state at <|emb_token|> back to a continuous embedding for nearest-neighbor retrieval. This hybrid is what lets one language model generate product identifiers autoregressively while still discriminating between products that share a discrete bucket. It is supported by a two-stage continual pre-training scheme —

Load-bearing premise

The load-bearing premise is that the online A/B numbers are statistically valid and causally attributable to Pailitao-MMSearch; the paper gives no experiment design, bucket sizes, or confidence intervals despite asserting significance.

What would settle it

Re-run the two deployments on freshly split traffic with pre-registered bucket sizes and a significance threshold: if the PeiWoGuang-floor GMV lift does not replicate above noise, or if the combined +13.61% disappears when the retrieval (+3.67%) and ranking (+9.94%) experiments are analyzed as separate, causally independent tests, the central effectiveness claim fails. A simpler check: verify whether +13.61% is the arithmetic sum of two different experiments rather than a measured combined lift.

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

Share X Bluesky LinkedIn Reddit HN

If this is right

  • If correct, a single generative model can replace the patchwork of text, image, and voice retrieval specialists in e-commerce search, eliminating ad-hoc fusion heuristics.
  • End-to-end product generation means candidate retrieval and ranking can be unified, shrinking the multi-stage cascade into one autoregressive step.
  • The continuous embedding component makes generative retrieval viable at billion-product scale, since a three-level discrete code space alone cannot distinguish near-identical items that share codes.
  • On-policy distillation offers a route to adapt proprietary base models to specialized domains without access to the original pretraining corpus, mitigating catastrophic forgetting.
  • Difficulty-aware reasoning lets the same model use fast direct generation for routine queries and chain-of-thought for compound intents, a design with direct latency implications.

Where Pith is reading between the lines

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

  • The headline +13.61% GMV is an arithmetic sum of two separate floor-level deployments (+3.67% retrieval and +9.94% ranking), not a single experiment; platform-wide GMV lifts are +1.02% and +0.25%. A fair public reading is that the true platform-level effect is likely an order of magnitude smaller than the headline. (editorial inference)
  • Because the paper reports no confidence intervals, bucket sizes, or p-values, the online numbers should be treated as directional until replication; a randomized re-run with pre-registered metrics would settle it. (editorial inference)
  • The HybSID design could transfer beyond e-commerce: any catalog-scale generative retrieval task where discrete quantization collapses fine distinctions could pair discrete codes with a trainable continuous emission token. (editorial inference)
  • The instruct-following offline results, if validated online, would open new query types (e.g., 'matching pants for this top'), but the paper itself says that function is not yet online. (editorial inference)

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper proposes Pailitao-MMSearch, a vision-language model for e-commerce multimodal search built on Qwen/Qwen2-VL. It introduces HybSID, a product tokenization combining three residual-quantized semantic codes with a continuous embedding emitted via a special <|emb_token|>; a two-stage continual pre-training stage (domain injection plus on-policy distillation for language preservation); and a post-training pipeline with multi-task SFT, difficulty-aware CoT, GRPO with product-grounded rewards, and multi-expert OPD. The authors report deployment on Taobao's Pailitao platform with online A/B lifts up to +13.61% GMV and +8.21% transaction volume, plus offline evaluations of SID translation/recognition, user sequence prediction, general-language benchmarks, and instruction-following search. The manuscript is an early technical report: several experiments are explicitly deferred to later versions.

Significance. If the online results are valid and causally attributable to the model, this would be a strong industrial demonstration that a single generative VLM with hybrid discrete/continuous product IDs can serve product retrieval and ranking. The HybSID representation is a plausible and interesting extension of prior semantic-ID generative retrieval, and the two-stage CPT idea addresses a real forgetting problem. The paper also has commendable transparency in reporting limitations and pending results, but this transparency also means the current version does not yet substantiate several of its central claims.

major comments (4)
  1. [§4.2.1–§4.2.2, Tables 4–5] The online evidence is not statistically supported as reported. No confidence intervals, p-values, sample sizes, traffic-allocation ratios, experiment windows, or baseline-system description are given, despite the conclusion asserting statistically significant improvements. More importantly, the abstract's headline '+13.61% GMV' is not a measured single-experiment effect: it is the arithmetic sum of +3.67% (generative retrieval, Table 4) and +9.94% (generative ranking, Table 5), two separate deployments. Summing lifts assumes additivity, no interference, and no overlapping traffic, none of which is argued. The manuscript should either report a properly designed joint/layered experiment or clearly qualify the 13.61% as a combined arithmetic figure rather than an observed causal lift.
  2. [§4.5, Table 9; §4.4] Three of the four claimed contributions—hybrid reasoning, reinforcement learning with verifiable rewards, and multi-expert OPD—have no supporting quantitative results in this version. Section 4.5 states 'Results for hybrid reasoning, RL enhancement, and multi-expert OPD will be updated in later arXiv versions,' and Table 9(b) is empty. Similarly, §4.4 states the instruct-following multimodal search function is 'not online yet.' The post-training pipeline is therefore presented without evaluation, although it is a load-bearing part of the system description. The current paper is better described as a system/methodology report with pending experiments.
  3. [§3.3.2, Table 7] The OPD claim of 'language capability preservation' is only partially supported. After Stage 2, IFEval strict improves from 36.78 to 38.45, but the base model scores 54.5; MMLU-Redux remains at 37.05 vs. 44.6 for the base; C-Eval and INCLUDE remain well below base; and LiveCodeBench drops. The paper's wording—'improving instruction following while maintaining overall general-language performance'—is accurate, but this is a partial recovery, not preservation. Since OPD is a central design choice, the effect size and remaining gap should be discussed explicitly, and the claim in the introduction that OPD 'recovers' instruction-following capabilities is too strong.
  4. [§4.3, Eq. (4)/(7)] The offline e-commerce understanding and retrieval metrics are closed-loop on the paper's own HybSID representation: SID translation/recognition and end-to-end retrieval are defined relative to the same residual-quantized codes and continuous embeddings used in training. This measures internal consistency of the learned mapping more than externally validated product understanding. For example, End-to-end Retrieval@K requires only regenerating the precomputed SID and embedding of the target product, and ground-truth intent labels come from handcrafted behavioral rules (§3.4.1). The authors should provide external grounding—e.g., human relevance judgments, comparison against a non-HybSID retrieval baseline, or a held-out behavioral validation—before claiming that the model has acquired genuine e-commerce knowledge.
minor comments (5)
  1. [§4.5/Table 9] The manuscript repeatedly defers results to 'later arXiv versions'; such placeholders are reasonable for a preprint but should be explicitly marked as preliminary in the metadata and conclusion, and the abstract should not present the full system as complete.
  2. [References] The FinGPT reference [Yang et al., 2023] contains the literal text 'PLACEHOLDER–verify exact details before submission.' This must be fixed before any camera-ready version.
  3. [Abstract vs. Conclusion] Inconsistent user-scale figures: the abstract says 'tens of millions of daily active users,' while the conclusion says 'hundreds of millions of users.' Please reconcile.
  4. [Table 8] The 'Gemini3.1-pro' judge is used for SIF/LIF rates, but the judging prompt, sample size, and judge-agreement statistics are not provided. Also, Matching Product Search with CoT shows lower Recall@10 (0.8356) than without CoT (0.8544); this trade-off should be explained.
  5. [§4.3.2] Table 7 states base-model scores 'are taken from the official Qwen3 report,' while Stage 1/2 scores come from the authors' evaluation. If the evaluation setups differ, the comparison may be unfair; please confirm all rows use identical prompting/evaluation settings.

Circularity Check

1 steps flagged

OPD-preservation evidence is selected on the complete evaluation set, making that supporting claim self-confirming; the headline online A/B claim remains an external anchor, so no total circularity.

specific steps
  1. fitted input called prediction [Section 4.3.2 (General Language Abilities), Table 7 discussion]
    "We select the Stage 2 OPD model trained with LR1 × 10−6 based on the complete evaluation. Relative to the Stage 1 model, it improves instruction following, raising IFEval strict-prompt accuracy from 36.78 to 38.45 (+1.67 points), while the remaining benchmarks in Table 7 stay essentially unchanged (e.g., MMLU-Redux is flat and C-Eval and INCLUDE move by less than 0.1 points)."

    The learning rate is chosen by optimizing the same evaluation benchmarks that are then reported as evidence that OPD preserves general language ability. The +1.67 IFEval gain is therefore the selected best value over the complete evaluation set, not an out-of-sample prediction; the claim 'improves instruction following while maintaining overall general-language performance' is an artifact of the model-selection criterion rather than an independent test of Stage 2 OPD. This does not contaminate the online A/B results, so the circularity is partial.

full rationale

The paper's central claim—production GMV/transaction lifts from online A/B testing—is externally anchored and not derived from the model's own outputs, so the headline is not circular. The offline 'e-commerce domain understanding' suite does reuse the same HybSID codes/embeddings as training targets, but that is standard supervised evaluation on held-out data rather than a definitional reduction. The clearest circular step is the Stage 2 OPD checkpoint selection: the authors state that the model was chosen 'based on the complete evaluation' and then cite that same evaluation's IFEval improvement as evidence that OPD preserves instruction following. This is fitted-input-called-prediction for a supporting claim. The TaoSID2.0-MM-CF citation involves overlapping contributors, but it is an implementation dependency, not a load-bearing uniqueness argument, so I do not count it as circularity. The paper also omits significance details for the A/B lifts and defers ablations, but those are verification/correctness gaps, not circularity per se.

Axiom & Free-Parameter Ledger

5 free parameters · 6 axioms · 1 invented entities

The central method relies on several unverified, mostly proprietary assumptions: the external SID codebook, the internal vision encoder, behavior-log-derived labels, verifiable RL rewards, and unvalidated A/B attribution. Free parameters (loss, reward, and distillation weights) are not reported with sensitivities. No physical invented entities are proposed; HybSID is a technical representation with only internal evidence.

free parameters (5)
  • λ (HybSID embedding-loss weight, Eq. 7) = not reported
    Balances discrete SID cross-entropy against continuous embedding MSE; no sensitivity analysis or ablation on λ is given.
  • α_r, β_r (RL reward weights, Eq. 10) = not reported
    Trade off exact SID-match reward against embedding-cosine reward; no sensitivity analysis is shown.
  • μ (Stage-2 OPD distillation strength, Eq. 8) = not reported
    Controls how strongly the student is pulled toward the base model; chosen via full-evaluation selection in §4.3.2.
  • w_k (multi-expert distillation weights, Eq. 11) = not reported
    Weights per expert in final OPD fusion; no ablation is provided because §4.5(b) defers the study.
  • difficulty threshold for normal vs hard SFT samples = not reported
    Determines which queries get direct generation vs chain-of-thought; based on heuristic semantic-distance rules in §3.4.1 with no calibration details.
axioms (6)
  • domain assumption The TaoSID2.0-MM-CF codebooks provide stable, meaningful semantic IDs for the billion-scale product catalog.
    §3.2.1 inherits the codebook (L=3, K=8192) without construction or validation details; a bad codebook would make HybSID generation meaningless.
  • domain assumption The frozen Pailitao vision encoder, trained on tens of billions of internal visual search data, produces query/product representations aligned with user visual intent.
    §3.1 introduces it without public description; central multimodal alignment depends on it.
  • ad hoc to paper User behavior co-occurrence (clicks, purchases) can be converted into ground-truth product-search intents via handcrafted rules.
    §3.4.1 Step 1 derives training labels from sequential user behaviors (e.g., cup then cup sleeve); no validation that these labels reflect search intent.
  • domain assumption The RL reward in Eq. (10), SID match plus embedding cosine, is verifiable and immune to reward hacking.
    §3.4.3 asserts the multi-reference reward mitigates exposure bias and reward hacking, but no reward-hacking analysis is supplied.
  • domain assumption On-policy distillation from the base LLM restores general capabilities without eroding injected e-commerce knowledge.
    §3.3.2 Eq. (8); Table 7 shows only partial recovery, so the assumption is only weakly supported.
  • domain assumption The online A/B setup isolates the effect of Pailitao-MMSearch from other platform changes and reports the chosen metrics without selection bias.
    §4.2 Tables 4 and 5 provide no design details; the central headline rests on this.
invented entities (1)
  • HybSID (3 discrete SID codes + <|emb_token|> continuous embedding) no independent evidence
    purpose: Product representation enabling the LLM to generate coarse SID codes and a fine continuous embedding end-to-end.
    The only evidence is the paper's internal offline and online evaluation; no code, weights, or third-party replication is available, and codebooks come from the same group's related system.

pith-pipeline@v1.3.0-alltime-deepseek · 19032 in / 18659 out tokens · 158256 ms · 2026-08-01T17:47:30.074899+00:00 · methodology

0 comments
Cite this review

Pith. "Pith review of Pailitao-MMSearch: Building Native E-Commerce Multimodal Search Foundation." pith.science (2026). https://pith.science/paper/4MYWYXK4

@misc{pith2026260717499,
  author       = {Pith},
  title        = {Pith review of: Pailitao-MMSearch: Building Native E-Commerce Multimodal Search Foundation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/4MYWYXK4}},
  note         = {Machine review of arXiv:2607.17499}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

The evolution of e-commerce has fundamentally transformed how users search for products, shifting from simple text-based keyword queries to complex multimodal interactions that seamlessly combine product images, natural language descriptions, and mixed-intent instructions. However, existing approaches face a critical dilemma: single-modal specialist models, deployed independently for text retrieval, visual search, and voice recognition, operate in isolation and cannot handle cross-modal queries, while general-purpose vision-language models lack the domain-specific knowledge necessary for fine-grained product understanding, user behavior modeling, and commercial intent reasoning. In this work, we present Pailitao-MMSearch, one native e-commerce multimodal search foundation model designed to bridge this gap. Our approach introduces three key innovations: (1)HybSID (Hybrid Semantic ID);(2)a two-stage continual pre-training strategy; and (3)a hybrid reasoning post-training pipeline. Built upon Qwen and deployed on Taobao's Pailitao multimodal search platform, Pailitao-MMSearch achieves substantial improvements in online A/B testing, including up to +13.61\% in Gross Merchandise Volume (GMV) and +8.21\% in transaction volume compared to traditional multi-modal search pipeline, demonstrating the effectiveness of our native e-commerce multimodal search large language models.

Figures

Figures reproduced from arXiv: 2607.17499 by Baicheng Chen, Bo Zheng, Chong Ma, Jian Ding, Jingqian Zhao, Jinsong Lan, Jun Song, Lianyu Du, Shuaiqi Jia, Shuwen Xiao, Xiangheng Kong, Xiaohan Ye, Xiaoyong Zhu, Xu Chen, Yuan Gao, Yunmeng Shu, Zhixiang Zhao, Zihan Gong.

Figure 1
Figure 1. Figure 1: Overall framework of Pailitao-MMSearch. The system builds upon a vision-language foun [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Architecture of HybSID. Each product is represented by three discrete SID tokens obtained [PITH_FULL_IMAGE:figures/full_fig_p010_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Two-stage continual pre-training pipeline. Stage 1 injects e-commerce domain knowledge [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Post-training pipeline overview. The CPT model undergoes multi-task SFT with difficulty [PITH_FULL_IMAGE:figures/full_fig_p012_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Real-world examples of our generative retrieval and ranking pipeline based on Pailitao [PITH_FULL_IMAGE:figures/full_fig_p017_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Similar Product Search: User uploads query image and free text, and the model generates [PITH_FULL_IMAGE:figures/full_fig_p020_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Matching Product Search: User uploads query image and free text, and the model gen [PITH_FULL_IMAGE:figures/full_fig_p021_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Matching Product Search with think mode: User uploads query image and free text, and [PITH_FULL_IMAGE:figures/full_fig_p022_8.png] view at source ↗

discussion (0)

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

Reference graph

Works this paper leans on

67 extracted references · 21 linked inside Pith

  1. [1]

    Item S age: Learning product embeddings for shopping recommendations at P interest

    Paul Baltescu, Haoyu Chen, Nikil Pancha, Andrew Zhai, and Charles Rosenberg. Item S age: Learning product embeddings for shopping recommendations at P interest. arXiv preprint arXiv:2205.11728, 2022

    Pith/arXiv arXiv 2022

  2. [2]

    Onesearch: A preliminary exploration of the unified end-to-end generative framework for e-commerce search, 2025

    Ben Chen, Xian Guo, Siyuan Wang, Zihan Liang, Yue Lv, Yufei Ma, Xinlong Xiao, Bowen Xue, Xuxin Zhang, Ying Yang, Huangyu Dai, Xing Xu, Tong Zhao, Mingcan Peng, Xiaoyang Zheng, Chao Wang, Qihang Zhao, Zhixin Zhai, Yang Zhao, Bochao Liu, Jingshan Lv, Xiao Liang, Yuqing Ding, Jing Chen, Chenyi Lei, Wenwu Ou, Han Li, and Kun Gai. Onesearch: A preliminary expl...

    arXiv 2025

  3. [3]

    Meditron-70b: Scaling medical pretraining for large language models, 2023 a

    Zeming Chen, Alejandro Hernández Cano, Angelika Romanou, Antoine Bonnet, Kyle Matoba, Francesco Salvi, Matteo Pagliardini, Simin Fan, Andreas Köpf, Amirkeivan Mohtashami, Alexandre Sallinen, Alireza Sakhaeirad, Vinitra Swamy, Igor Krawczuk, Deniz Bayazit, Axel Marmet, Syrielle Montariol, Mary-Anne Hartley, Martin Jaggi, and Antoine Bosselut. Meditron-70b:...

    Pith/arXiv arXiv 2023

  4. [4]

    Intern VL : Scaling up vision foundation models and aligning for generic visual-linguistic tasks

    Zhe Chen, Jiannan Wu, Wenhai Wang, Weijie Su, Guo Chen, Sen Xing, Muyan Zhong, Qinglong Zhang, Xizhou Zhu, Lewei Lu, Bin Li, Ping Luo, Tong Lu, Yu Qiao, and Jifeng Dai. Intern VL : Scaling up vision foundation models and aligning for generic visual-linguistic tasks. arXiv preprint arXiv:2312.14238, 2023 b

    Pith/arXiv arXiv 2023

  5. [6]

    Forge: Forming semantic identifiers for generative retrieval in industrial datasets, 2026

    Kairui Fu, Tao Zhang, Shuwen Xiao, Ziyang Wang, Xinming Zhang, Chenchi Zhang, Yuliang Yan, Junjun Zheng, Xiangheng Kong, Shengyu Zhang, Kun Kuang, and Yuning Jiang. Forge: Forming semantic identifiers for generative retrieval in industrial datasets, 2026. URL https://arxiv.org/abs/2509.20904

    Pith/arXiv arXiv 2026

  6. [7]

    Sim CSE : Simple contrastive learning of sentence embeddings

    Tianyu Gao, Xingcheng Yao, and Danqi Chen. Sim CSE : Simple contrastive learning of sentence embeddings. arXiv preprint arXiv:2104.08821, 2021

    Pith/arXiv arXiv 2021

  7. [8]

    Recommendation as language processing ( RLP ): A unified pretrain, personalized prompt & predict paradigm ( P5 )

    Shijie Geng, Shuchang Liu, Zuohui Fu, Yingqiang Ge, and Yongfeng Zhang. Recommendation as language processing ( RLP ): A unified pretrain, personalized prompt & predict paradigm ( P5 ). Proceedings of the 16th ACM Conference on Recommender Systems, 2022

    2022

  8. [9]

    Learning deep structured semantic models for web search using clickthrough data

    Po-Sen Huang, Xiaodong He, Jianfeng Gao, Li Deng, Alex Acero, and Larry Heck. Learning deep structured semantic models for web search using clickthrough data. In Proceedings of the 22nd ACM International Conference on Information & Knowledge Management, pages 2333--2338, 2013. doi:10.1145/2505515.2505665

    arXiv 2013

  9. [11]

    Rusu, Kieran Milan, John Quan, Tiago Ramalho, Agnieszka Grabska-Barwinska, et al

    James Kirkpatrick, Razvan Pascanu, Neil Rabinowitz, Joel Veness, Guillaume Desjardins, Andrei A. Rusu, Kieran Milan, John Quan, Tiago Ramalho, Agnieszka Grabska-Barwinska, et al. Overcoming catastrophic forgetting in neural networks. Proceedings of the National Academy of Sciences, 114 0 (13): 0 3521--3526, 2017

    2017

  10. [12]

    Autoregressive image generation using residual quantization

    Doyup Lee, Chiheon Kim, Saehoon Kim, Minsu Cho, and Wook-Shin Han. Autoregressive image generation using residual quantization. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022

    2022

  11. [13]

    Visual instruction tuning

    Haotian Liu, Chunyuan Li, Qingyang Wu, and Yong Jae Lee. Visual instruction tuning. Advances in Neural Information Processing Systems, 36, 2024

    2024

  12. [14]

    An empirical study of catastrophic forgetting in large language models during continual fine-tuning, 2025

    Yun Luo, Zhen Yang, Fandong Meng, Yafu Li, Jie Zhou, and Yue Zhang. An empirical study of catastrophic forgetting in large language models during continual fine-tuning, 2025. URL https://arxiv.org/abs/2308.08747

    Pith/arXiv arXiv 2025

  13. [16]

    Michael McCloskey and Neal J. Cohen. Catastrophic interference in connectionist networks: The sequential learning problem. Psychology of Learning and Motivation, 24: 0 109--165, 1989

    1989

  14. [17]

    GPT-4 technical report

    OpenAI, Josh Achiam, Steven Adler, Sandhini Agarwal, et al. GPT-4 technical report. arXiv preprint arXiv:2303.08774, 2023

    Pith/arXiv arXiv 2023

  15. [18]

    Keshavan, Trung Vu, Lukasz Heldt, Lichan Hong, Yi Tay, Vinh Q

    Shashank Rajput, Nikhil Mehta, Anima Singh, Raghunandan H. Keshavan, Trung Vu, Lukasz Heldt, Lichan Hong, Yi Tay, Vinh Q. Tran, Jonah Samost, Maciej Kula, Ed H. Chi, and Maheswaran Sathiamoorthy. Recommender systems with generative retrieval. Advances in Neural Information Processing Systems, 36, 2023

    2023

  16. [21]

    Tran, Mostafa Dehghani, Jianmo Ni, Dara Bahri, Harsh Mehta, Zhen Qin, Kai Hui, Zhe Zhao, Jai Gupta, Tal Schuster, William W

    Yi Tay, Vinh Q. Tran, Mostafa Dehghani, Jianmo Ni, Dara Bahri, Harsh Mehta, Zhen Qin, Kai Hui, Zhe Zhao, Jai Gupta, Tal Schuster, William W. Cohen, and Donald Metzler. Transformer memory as a differentiable search index. Advances in Neural Information Processing Systems, 35, 2022

    2022

  17. [24]

    Qwen2- VL : Enhancing vision-language model's perception of the world at any resolution

    Peng Wang, Shuai Bai, Sinan Tan, Shijie Wang, Zhihao Fan, Jinze Bai, Keqin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, Yang Fan, Kai Dang, Mengfei Du, Xuancheng Ren, Rui Men, Dayiheng Liu, Chang Zhou, Jingren Zhou, and Junyang Lin. Qwen2- VL : Enhancing vision-language model's perception of the world at any resolution. arXiv preprint arXiv:2409.12191, 2024

    Pith/arXiv arXiv 2024

  18. [25]

    Pumgpt: A large vision-language model for product understanding, 2024

    Wei Xue, Zongyi Guo, Baoliang Cui, Zheng Xing, Xiaoyi Zeng, Xiufei Wang, Shuhui Wu, and Weiming Lu. Pumgpt: A large vision-language model for product understanding, 2024. URL https://arxiv.org/abs/2308.09568

    Pith/arXiv arXiv 2024

  19. [27]

    Visual search at eBay

    Fan Yang, Ajinkya Kale, Yury Bubnov, Leon Stein, Qiaosong Wang, Hadi Kiapour, and Robinson Piramuthu. Visual search at eBay . arXiv preprint arXiv:1706.03154, 2017

    Pith/arXiv arXiv 2017

  20. [28]

    Fin GPT : Open-source financial large language models

    Hongyang Yang, Xiao-Yang Liu, and Christina Dan Wang. Fin GPT : Open-source financial large language models. arXiv preprint arXiv:2306.06031, 2023. PLACEHOLDER -- verify exact details before submission

    arXiv 2023

  21. [29]

    e P roduct: A million-scale visual search benchmark to address product recognition challenges

    Jiangbo Yuan, An-Ti Chiang, Wen Tang, and Antonio Haro. e P roduct: A million-scale visual search benchmark to address product recognition challenges. arXiv preprint arXiv:2107.05856, 2021

    Pith/arXiv arXiv 2021

  22. [30]

    Learning a unified embedding for visual search at Pinterest

    Andrew Zhai, Hao-Yu Wu, Eric Tzeng, Dong Huk Park, and Charles Rosenberg. Learning a unified embedding for visual search at Pinterest . In Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, 2019

    2019

  23. [31]

    Recommendation as instruction following: A large language model empowered recommendation approach, 2023

    Junjie Zhang, Ruobing Xie, Yupeng Hou, Wayne Xin Zhao, Leyu Lin, and Ji-Rong Wen. Recommendation as instruction following: A large language model empowered recommendation approach, 2023. URL https://arxiv.org/abs/2305.07001

    Pith/arXiv arXiv 2023

  24. [32]

    Adapting large language models by integrating collaborative semantics for recommendation

    Bowen Zheng, Yupeng Hou, Hongyu Lu, Yu Chen, Wayne Xin Zhao, Ming Chen, and Ji-Rong Wen. Adapting large language models by integrating collaborative semantics for recommendation. In 2024 IEEE 40th International Conference on Data Engineering (ICDE), pages 1435--1448. IEEE, 2024

    2024

  25. [33]

    Deep interest network for click-through rate prediction

    Guorui Zhou, Chengru Song, Xiaoqiang Zhu, Ying Fan, Han Zhu, Xiao Ma, Yanghui Yan, Junqi Jin, Han Li, and Kun Gai. Deep interest network for click-through rate prediction. In Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, 2018

    2018

  26. [34]

    Onerec-v2 technical report, 2025

    Guorui Zhou, Hengrui Hu, Hongtao Cheng, Huanjie Wang, Jiaxin Deng, Jinghao Zhang, Kuo Cai, Lejian Ren, Lu Ren, Liao Yu, Pengfei Zheng, Qiang Luo, Qianqian Wang, Qigen Hu, Rui Huang, Ruiming Tang, Shiyao Wang, Shujie Yang, Tao Wu, Wuchao Li, Xinchen Luo, Xingmei Wang, Yi Su, Yunfan Wu, Zexuan Cheng, Zhanyu Liu, Zixing Zhang, Bin Zhang, Boxuan Wang, Chaoyi ...

    Pith/arXiv arXiv 2025

  27. [35]

    Learning tree-based deep model for recommender systems

    Han Zhu, Xiang Li, Pengye Zhang, Guozheng Li, Jie He, Han Li, and Kun Gai. Learning tree-based deep model for recommender systems. In Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, pages 1079--1088, 2018. doi:10.1145/3219819.3219826

    arXiv 2018

  28. [36]

    Wang, Peng and Bai, Shuai and Tan, Sinan and Wang, Shijie and Fan, Zhihao and Bai, Jinze and Chen, Keqin and Liu, Xuejing and Wang, Jialin and Ge, Wenbin and Fan, Yang and Dang, Kai and Du, Mengfei and Ren, Xuancheng and Men, Rui and Liu, Dayiheng and Zhou, Chang and Zhou, Jingren and Lin, Junyang , journal=. Qwen2-

  29. [37]

    arXiv preprint arXiv:2412.15115 , year=

    Qwen2.5 Technical Report , author=. arXiv preprint arXiv:2412.15115 , year=

    Pith/arXiv arXiv

  30. [38]

    Advances in Neural Information Processing Systems , volume=

    Visual Instruction Tuning , author=. Advances in Neural Information Processing Systems , volume=

  31. [39]

    OpenAI and Achiam, Josh and Adler, Steven and Agarwal, Sandhini and others , journal=

  32. [40]

    Chen, Zhe and Wu, Jiannan and Wang, Wenhai and Su, Weijie and Chen, Guo and Xing, Sen and Zhong, Muyan and Zhang, Qinglong and Zhu, Xizhou and Lu, Lewei and Li, Bin and Luo, Ping and Lu, Tong and Qiao, Yu and Dai, Jifeng , journal=. Intern

  33. [41]

    Advances in Neural Information Processing Systems , volume=

    Recommender Systems with Generative Retrieval , author=. Advances in Neural Information Processing Systems , volume=

  34. [42]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , year=

    Autoregressive Image Generation using Residual Quantization , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , year=

  35. [43]

    Advances in Neural Information Processing Systems , volume=

    Transformer Memory as a Differentiable Search Index , author=. Advances in Neural Information Processing Systems , volume=

  36. [44]

    arXiv preprint arXiv:2010.00904 , year=

    Autoregressive Entity Retrieval , author=. arXiv preprint arXiv:2010.00904 , year=

    Pith/arXiv arXiv 2010

  37. [45]

    Advances in Neural Information Processing Systems , volume=

    Training language models to follow instructions with human feedback , author=. Advances in Neural Information Processing Systems , volume=

  38. [46]

    Advances in Neural Information Processing Systems , volume=

    Direct Preference Optimization: Your Language Model is Secretly a Reward Model , author=. Advances in Neural Information Processing Systems , volume=

  39. [47]

    arXiv preprint arXiv:2402.03300 , year=

    DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models , author=. arXiv preprint arXiv:2402.03300 , year=

    Pith/arXiv arXiv

  40. [48]

    Recommendation as Language Processing (

    Geng, Shijie and Liu, Shuchang and Fu, Zuohui and Ge, Yingqiang and Zhang, Yongfeng , journal=. Recommendation as Language Processing (

  41. [49]

    arXiv preprint arXiv:2308.12950 , year=

    Code Llama: Open Foundation Models for Code , author=. arXiv preprint arXiv:2308.12950 , year=

    Pith/arXiv arXiv

  42. [50]

    Proceedings of the National Academy of Sciences , volume=

    Overcoming catastrophic forgetting in neural networks , author=. Proceedings of the National Academy of Sciences , volume=. 2017 , publisher=

    2017

  43. [51]

    Psychology of Learning and Motivation , volume=

    Catastrophic interference in connectionist networks: The sequential learning problem , author=. Psychology of Learning and Motivation , volume=. 1989 , publisher=

    1989

  44. [52]

    2025 , eprint=

    An Empirical Study of Catastrophic Forgetting in Large Language Models During Continual Fine-tuning , author=. 2025 , eprint=

    2025

  45. [53]

    2024 IEEE 40th International Conference on Data Engineering (ICDE) , pages=

    Adapting large language models by integrating collaborative semantics for recommendation , author=. 2024 IEEE 40th International Conference on Data Engineering (ICDE) , pages=. 2024 , organization=

    2024

  46. [54]

    2023 , eprint=

    Recommendation as Instruction Following: A Large Language Model Empowered Recommendation Approach , author=. 2023 , eprint=

    2023

  47. [55]

    2024 , isbn =

    Ma, Heli and Zou, Jie and Aliannejadi, Mohammad and Kanoulas, Evangelos and Bin, Yi and Yang, Yang , title =. 2024 , isbn =. doi:10.1145/3627673.3679875 , booktitle =

    arXiv 2024

  48. [56]

    2024 , eprint=

    PUMGPT: A Large Vision-Language Model for Product Understanding , author=. 2024 , eprint=

    2024

  49. [57]

    2023 , eprint=

    MEDITRON-70B: Scaling Medical Pretraining for Large Language Models , author=. 2023 , eprint=

    2023

  50. [58]

    Yang, Hongyang and Liu, Xiao-Yang and Wang, Christina Dan , journal=. Fin. 2023 , note=

    2023

  51. [59]

    arXiv preprint arXiv:2505.09388 , year=

    Qwen3 Technical Report , author=. arXiv preprint arXiv:2505.09388 , year=

    Pith/arXiv arXiv

  52. [60]

    Proceedings of the 22nd ACM International Conference on Information & Knowledge Management , pages=

    Learning Deep Structured Semantic Models for Web Search using Clickthrough Data , author=. Proceedings of the 22nd ACM International Conference on Information & Knowledge Management , pages=. 2013 , doi=

    2013

  53. [61]

    Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining , pages=

    Learning Tree-based Deep Model for Recommender Systems , author=. Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining , pages=. 2018 , doi=

    2018

  54. [62]

    Visual Search at

    Yang, Fan and Kale, Ajinkya and Bubnov, Yury and Stein, Leon and Wang, Qiaosong and Kiapour, Hadi and Piramuthu, Robinson , journal=. Visual Search at

  55. [63]

    Learning a Unified Embedding for Visual Search at

    Zhai, Andrew and Wu, Hao-Yu and Tzeng, Eric and Park, Dong Huk and Rosenberg, Charles , booktitle=. Learning a Unified Embedding for Visual Search at

  56. [64]

    Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining , year=

    Deep Interest Network for Click-Through Rate Prediction , author=. Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining , year=

  57. [65]

    arXiv preprint arXiv:2103.00020 , year=

    Learning Transferable Visual Models From Natural Language Supervision , author=. arXiv preprint arXiv:2103.00020 , year=

    Pith/arXiv arXiv

  58. [66]

    2026 , eprint=

    Pailitao-VL: Unified Embedding and Reranker for Real-Time Multi-Modal Industrial Search , author=. 2026 , eprint=

    2026

  59. [67]

    2025 , eprint=

    ELIP: Enhanced Visual-Language Foundation Models for Image Retrieval , author=. 2025 , eprint=

    2025

  60. [68]

    arXiv preprint arXiv:2004.04906 , year=

    Dense Passage Retrieval for Open-Domain Question Answering , author=. arXiv preprint arXiv:2004.04906 , year=

    Pith/arXiv arXiv 2004

  61. [69]

    Gao, Tianyu and Yao, Xingcheng and Chen, Danqi , journal=. Sim

  62. [70]

    arXiv preprint arXiv:2212.03533 , year=

    Text Embeddings by Weakly-Supervised Contrastive Pre-training , author=. arXiv preprint arXiv:2212.03533 , year=

    Pith/arXiv arXiv

  63. [71]

    Baltescu, Paul and Chen, Haoyu and Pancha, Nikil and Zhai, Andrew and Rosenberg, Charles , journal=. Item

  64. [72]

    Yuan, Jiangbo and Chiang, An-Ti and Tang, Wen and Haro, Antonio , journal=. e

  65. [73]

    2025 , eprint=

    OneSearch: A Preliminary Exploration of the Unified End-to-End Generative Framework for E-commerce Search , author=. 2025 , eprint=

    2025

  66. [74]

    2025 , eprint=

    OneRec-V2 Technical Report , author=. 2025 , eprint=

    2025

  67. [75]

    2026 , eprint=

    FORGE: Forming Semantic Identifiers for Generative Retrieval in Industrial Datasets , author=. 2026 , eprint=

    2026