Pith. sign in

REVIEW 4 major objections 6 minor 40 references

Traceable LLM Reasoning for Fake-Order Fraud Detection

T0 review · 4 major / 6 minor · reviewed 2026-08-01 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read DeepScrub shows that a small, domain-tuned LLM with suggest-reflect reinforcement learning can detect fake-order fraud more accurately than a much larger general model or routine human reviewers, while producing inspectable reasoning traces

desk verdict A plausible and well-engineered fraud-detection system whose central traceability claim is not actually measured; worth refereeing if the authors can supply a real faithfulness check. read the letter →

arxiv 2607.23075 v1 pith:3MKVM7BG submitted 2026-07-25 cs.CR cs.AIcs.LG

Siqi You , Bingsong Xu , Zhixian Zheng , Xinjian Peng , Yang Xie , Ying Wang , Jiarong Xu This is my paper
classification cs.CRcs.AIcs.LG
keywords fake-orderfraudbrushingdetectionLLMreasoningreinforcementlearningtraceableevidencesuggest-reflectO2Oplatformrisk-controlreview
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper sets out to prove that fake-order fraud detection on large online-to-offline platforms can be made simultaneously more accurate and more auditable by replacing handcrafted features and black-box classifiers with a compact language model that reasons in text. The proposed system, DeepScrub, translates graphs, behavioral sequences, and transaction tables into unified text, continues pretraining on risk-control material, and then trains the model with a suggest-reflect reinforcement loop in which the model either double-checks its own reasoning or receives targeted suggestions from expert models before giving a final verdict. Trained this way, an 8B-parameter model reaches 85.3% macro-F1 offline, beating a 32B general model and all fine-tuned baselines. In a four-week live pilot it reached 91.8% precision and 88.5% recall, improving on routine first-stage human reviewers by 16.6 and 38.8 percentage points, while cutting first-stage manual review workload by 94% and saving nearly one million RMB per year. The payoff the authors argue for is a deployed reviewer whose verdicts come with reasoning chains that enforcement and appeal workflows can inspect.

What carries the argument

The load-bearing mechanism is SURE (SUggest-REflect), a two-round reinforcement training loop. The policy first samples a group of candidate answers; a dual-branch suggestion module decides, from the accuracy reward, whether a candidate is reliable enough for intrinsic self-verification or needs extrinsic hints from a transaction-expert model and a risk-control expert model; the policy then re-reasons with those suggestions to emit a refined answer. The advantages of the first- and second-round responses are blended into a mixed advantage used in a group-relative policy optimization update. Three rewards shape the behavior: a format reward enforcing structured <reason>/<answer> output; an ac

What would settle it

Take a sample of orders the model classified, and for each one remove or contradict the specific evidence named in its reasoning trace (for example, 'abnormal surge in transaction amount' or '17 linked accounts') while leaving the rest of the input intact; if the verdict stays the same in a large share of cases, the traces are post-hoc rationalizations rather than decision evidence. A second check: replace the reference reasoning used during training with a deliberately inconsistent rationale and confirm the reasoning reward decreases — if it cannot detect the inconsistency, the traceability m

Watch

Extended reading notes

Core claim

DeepScrub's central claim is that jointly optimizing prediction correctness and reasoning quality produces a fraud reviewer that is more accurate and more inspectable than rule-based screening, black-box models, or routine human first-stage review. The framework unifies heterogeneous risk evidence into text, continues pretraining on a mix of risk-control and general corpora, and trains with a SUggest-REflect (SURE) mechanism: low-confidence first-round answers receive suggestions from transaction- and risk-control expert models before the policy reflects and produces a final answer, while high-confidence answers are prompted to self-verify. The advantages of both rounds are blended and appli

Load-bearing premise

The load-bearing premise is that the reasoning traces the model is trained and rewarded to emit genuinely reflect the evidence driving its decisions, rather than being post-hoc narratives produced after the label is known; the paper does not independently verify this faithfulness.

Editorial extensions

If this is right

  • Small, task-optimized LLMs can beat much larger general-purpose LLMs on specialized risk-control tasks, so deployment cost need not scale with model size.
  • Fraud verdicts can be delivered with human-readable reasoning chains usable in enforcement, appeals, and expert adjudication.
  • Automating first-stage review at 91.8% precision and 88.5% recall can reduce manual review workload by 94% and save roughly one million RMB per year on a large O2O platform.
  • The suggest-reflect training recipe is transferable to other high-stakes classification settings where both accuracy and explainability are required.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The auditability benefit stands or falls on whether the emitted rationales faithfully reflect the evidence that drives the decision; the paper rewards similarity to reference rationales generated by a teacher model but does not independently verify that the traces are causally tied to the prediction. A perturbation test — deleting the cited evidence from an input and checking whether the predictio
  • The same semantic-unification-plus-suggest-reflect recipe could generalize to other domains with heterogeneous features and defensibility requirements, such as credit underwriting, content moderation, or insurance claim review.
  • The pilot compares against routine first-stage reviewers, not senior fraud experts; reported human-beating margins should be read against that baseline, and an expert-level comparison would be a stricter test of the model's value.
  • The found 1:15 ratio of domain to general pretraining data suggests a tunable trade-off between domain fluency and general capability; other deployments would need to retune it for their data mix.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. DeepScrub is an LLM-based reinforcement learning framework for fake-order fraud detection in O2O platforms. It converts heterogeneous risk signals (graphs, sequences, tables) into textual evidence via semantic unification, continues pretraining on risk-control and general corpora, and trains a Qwen3-8B policy with SURE, a two-round suggest-reflect RL procedure with format, accuracy, and reasoning rewards. The model outputs a label, confidence, and reasoning trace. On a 150K-order proprietary dataset, DeepScrub reports macro-F1 of 85.3%, outperforming fine-tuned 8B baselines and a zero-shot 32B model. In a four-week live pilot, it reports 91.8% precision and 88.5% recall versus 75.2% and 49.7% for first-stage human reviewers, with a 94% workload reduction. The paper's stated contribution is improved detection plus traceable reasoning for enforcement and appeal workflows.

Significance. If the traceability claim were supported, this would be a practically valuable contribution: it combines domain-adapted LLMs with RL for a high-stakes fraud-review task, includes a real-world deployment pilot, and shows that a task-tuned 8B model can outperform a larger general model. The ablation design and chronological split are sensible, and the pilot provides concrete operational evidence. However, the paper's distinctive claim—that the reasoning traces are auditable and faithful—is not yet substantiated. The rationales used as training targets are themselves generated from ground-truth labels, and no external faithfulness measure is presented. The reported performance gains also lack variance estimates. Because the auditability motivation is central and currently rests on qualitative examples and a reward function with a conditioning error (Eq. 5), the paper requires major revision rather than acceptance.

major comments (4)
  1. [Sec. 4.2; Eqs. (4), (10)] The traceable-reasoning claim is trained and evaluated against label-conditioned rationales. CoT Generation (Sec. 4.2, App. A.4) asks DeepSeek-R1 to produce a rationale with the true label as input; the only filter is label consistency. The accuracy and reasoning rewards (Eqs. 4, 10) then reward similarity to these same rationales. A model can therefore receive high reasoning rewards by reproducing label-consistent narrative templates without grounding its decision in the transaction evidence. Tables 3-4 are qualitative case studies and do not measure faithfulness. To support the core auditability contribution, the paper needs an external test: e.g., human auditors rating whether each trace's cited evidence is present and sufficient, or counterfactual tests in which evidence is altered and the rationale changes accordingly.
  2. [Sec. 4.3, Eq. (5)] Eq. (5) defines A(ê,y) = (1/|y|) Σ_n Pθ(y_n | concat(ê, y_<n)); the input prompt x (the transaction evidence) does not appear in the conditioning set. Consequently, the 'reasoning reward' measures only local coherence between the generated reasoning ê and the answer y, not whether ê is grounded in the transaction data. The formula should be Pθ(y_n | x, concat(ê, y_<n)). As written, Δscore in Eq. (8) can reward a plausible-sounding rationale that ignores the evidence—exactly the failure mode the paper claims to avoid.
  3. [Sec. 6.2-6.3, Table 2] All offline results are single-run point estimates with no error bars, significance tests, or multiple seeds. The headline gain over the strongest fine-tuned baseline is 2.7 macro-F1 points (0.853 vs 0.826), and the w/o r_reasoning ablation is only 0.9 points lower. Without variance estimates, it is unclear whether these differences are systematic. Please report means and standard deviations over at least three seeds, and sensitivity to the main hyperparameters α, μ, ω (Secs. 4.3, 6.1). This is needed to support the central performance and ablation claims.
  4. [Sec. 7] The pilot's human baseline is described only as 'standard first-stage reviewers' with routine training; their workload, selection criteria, and review context are not specified. Final labels are adjudicated by senior experts, but the paper does not report inter-annotator agreement, the number of cases escalated, or the escalation threshold. The 94% workload reduction is presented without accounting for the cost of expert adjudication, downstream false-negative review, or inference/compute costs. These details are necessary to substantiate the production-impact and cost-savings claims.
minor comments (6)
  1. [Sec. 4.3, Eq. (1)] The confidence weights 1, 2/3, 1/3 are arbitrary; no justification or sensitivity analysis is provided.
  2. [Sec. 6.1] Training details are incomplete: the SFT portion (20% of the CoT-augmented set) has no data size, number of epochs, or early-stopping criterion; the format reward is also never formally defined.
  3. [Fig. 2] The arrow 'without any annotation CoT generation' is misleading, since the CoT Generation prompt in App. A.4 includes the true label. Suggest rewording to 'without a manual rationale'.
  4. [Table 1] No evaluation details for C-Eval/C-MMLU (few-shot setup, prompt) are given; please add.
  5. [Abstract/Conclusion] The 'nearly one million RMB annually' saving is stated without a calculation; a short cost model or footnote would help.
  6. [App. A.1] Given the reproducibility claim, the prompts and semantic templates should be released even if the proprietary dataset cannot be.

Circularity Check

1 steps flagged · score 5.0 of 10

Traceable-reasoning claim is defined and evaluated against the same label-conditioned DeepSeek-R1 rationales used for training; detection-accuracy claim remains externally grounded.

  1. self definitional [Sec. 4.2 'CoT Data Amplification'; Eqs. (3), (4), (9), (10); Sec. 6.4]
    "Orders without rationales are processed through CoT generation conditioned on their labels. We then filter generated rationales through manual verification to reduce label-inconsistent reasoning. ... we further introduce a process-level score defined as the textual similarity between the model-generated reasoning ˆe and the ground-truth reasoning e ... r_accuracy = (1−α)·I(ˆy,y)·w(ˆc)+α·sim(ˆe,e). ... r_reason = Δscore×γ."

    The paper's 'reasoning quality' target e is not an independently verified evidence trace: for rationale-free orders, it is generated by DeepSeek-R1 conditioned on the label. Eq. (4) rewards the policy for matching those label-conditioned rationales, and Eq. (10) rewards overlap with keywords K(y) associated with the ground-truth answer. Any improvement measured by these rewards is therefore by construction an improvement in imitating the same synthetic rationales, not in evidence-grounded traceability. The only evaluation of traceable reasoning offered in Sec. 6.4 is qualitative (Tables 3–4); no external faithfulness metric is used. Thus the distinctive auditability claim is asserted through the very object the model was trained to reproduce, making that sub-claim partially self-definition

full rationale

The headline operational claims are not circular: the fraud labels come from enforcement outcomes, high-precision rules, law-enforcement confirmation, and senior-expert adjudication; the offline evaluation uses an external chronological split and baselines, and the pilot compares against human reviewers with expert-adjudicated labels. These parts of the derivation chain are grounded in outside data. The circularity concern is confined to the paper's distinctive 'traceable reasoning' contribution. In Sec. 4.2, CoT rationales for unlabeled orders are generated by DeepSeek-R1 conditioned on the label, filtered only for label consistency; Eqs. (4) and (10) then train the model to match those rationales and to include label-specific keywords. No independent check establishes that the resulting traces are causally grounded in the input evidence rather than post-hoc, label-consistent narratives. The qualitative case studies in Sec. 6.4 do not measure faithfulness against an external standard. So the reasoning-quality contribution is substantially self-referential, even though the detection-accuracy contribution is not. No load-bearing self-citation chain was found.

Assumptions & free parameters 6 free parameters · 6 assumptions · 0 invented entities

The central empirical claims rest on proprietary labels, expert adjudication, and the validity of LLM-generated rationales as ground truth for reasoning. No new physical or conceptual entities are introduced; the free parameters are reward and training hyperparameters that are not sensitivity-analyzed.

free parameters (6)
  • alpha (reward blend) = 0.1
    Weight of process-similarity term in accuracy reward, Eq. (4); no sensitivity analysis.
  • mu (SURE branch threshold) = 0.5
    Accuracy-reward threshold deciding intrinsic vs extrinsic suggestions, Eq. (11).
  • omega (advantage mixing) = 0.7
    Mixing weight between second-round advantage and advantage increment, Eq. (12).
  • confidence weights w(c) = 1, 2/3, 1/3
    Arbitrary mapping of high/medium/low confidence to reward weights, Eq. (1).
  • risk-control:general pretraining ratio = 1:15
    Chosen as best trade-off on C-Eval/C-MMLU in Table 1; not tied to the fraud task.
  • GRPO group size / LR / warmup / batch = G=6, 1e-5, 0.05, 16
    Training hyperparameters that influence the reported F1; no grid or sensitivity results.
assumptions (6)
  • domain assumption Enforcement outcomes, high-precision rules, and law-enforcement confirmations provide reliable positive labels.
    Sec. 5 uses these as the 'most reliable source of positive labels'; if these labels are noisy, all reported F1 scores are affected.
  • domain assumption Benign samples can be selected via 'high fraud cost or logically low fraud probability' heuristics.
    Sec. 5; mislabeled benign samples would inflate precision/recall on the normal class.
  • domain assumption De-identified textual conversion of graphs/sequences/tables preserves the evidence needed for correct fraud decisions.
    Sec. 4.2 assumes the semantic unification templates do not lose discriminative signal.
  • domain assumption DeepSeek-R1-augmented/generated CoT rationales, after manual verification, are valid ground-truth reasoning.
    Sec. 4.2 feeds these rationales into rewards (Eqs. 4, 10); if they are post-hoc or label-leaking, the reasoning-quality claim is unsupported.
  • domain assumption Senior expert adjudication in the pilot provides correct final labels for comparing human vs model recall.
    Sec. 7; the human-baseline comparison depends on this adjudication.
  • domain assumption LLM-generated reasoning traces faithfully reflect the decision process.
    The paper's 'traceable evidence' framing (Sec. 1, 7) assumes rationales are not post-hoc rationalizations; never tested.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Traceable LLM Reasoning for Fake-Order Fraud Detection." pith.science (2026). https://pith.science/paper/3MKVM7BG

@misc{pith2026260723075,
  author       = {Pith},
  title        = {Pith review of: Traceable LLM Reasoning for Fake-Order Fraud Detection},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3MKVM7BG}},
  note         = {Machine review of arXiv:2607.23075}
}
read the original abstract

Detecting fake-order fraud at scale remains a critical challenge for large online-to-offline (O2O) service platforms, as existing approaches often rely on expert-designed features, produce black-box decisions, and provide limited interpretability. To address these limitations, we propose DeepScrub, a reinforcement learning framework built upon large language models (LLMs) for fake-order fraud detection with traceable reasoning. DeepScrub introduces three innovations. First, a semantic unification module converts heterogeneous risk signals into textual descriptions that LLMs can understand. Second, continued pre-training on risk-control corpora injects domain knowledge, and task rewards jointly evaluate prediction correctness and reasoning quality. Third, the SUggest-REflect (SURE) mechanism incorporates expert feedback and model self-checking to iteratively refine reasoning paths. On a real-world fake-order fraud detection dataset, DeepScrub achieves a macro-F1 score of 85.3%, outperforming the best baseline by 2.7 percentage points. Our task-optimized 8B model further surpasses a 32B model, showing that domain adaptation can matter more than model scale in this setting. In a four-week live pilot, DeepScrub achieved 91.8% precision and 88.5% recall, improving over first-stage human reviewers by 16.6 and 38.8 percentage points. It reduced first-stage manual review workload by 94% and saved nearly one million RMB annually. These results show that DeepScrub improves fraud review accuracy, reduces first-stage review workload, and provides traceable evidence for production risk-review workflows.

Figures

Figures reproduced from arXiv: 2607.23075 by the authors.

Figure 1
Figure 1. Overview of DeepScrub. The system first adapts a base LLM with risk-control corpora, then converts heterogeneous [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Construction of reasoning-augmented training data. [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. SURE training mechanism. The policy model first generates candidate responses, receives intrinsic or expert suggestions [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

40 extracted references · 9 linked inside Pith

  1. [1]

    Abdulalem Ali, Shukor Abd Razak, Siti Hajar Othman, Taiseer Abdalla Elfadil Eisa, Arafat Al-Dhaqm, Maged Nasser, Tusneem Elhassan, Hashim Elshafie, and Abdu Saif. 2022. Financial fraud detection based on machine learning: a systematic literature review.Applied Sciences12, 19 (2022), 9637

    2022

  2. [2]

    Zheng Cai, Maosong Cao, Haojiong Chen, Kai Chen, Keyu Chen, et al . 2024. InternLM2 Technical Report. arXiv:2403.17297 [cs.CL]

    arXiv 2024

  3. [3]

    Dawei Cheng, Yao Zou, Sheng Xiang, and Changjun Jiang. 2025. Graph neural networks for financial fraud detection: a review.Frontiers of Computer Science19, 9 (2025), 199609

    2025

  4. [4]

    Sijie Cheng, Yanbo Yang, Jiawei Zhang, and Pengfei Li. 2025. LLM-TAD: Inter- pretable Ethereum Fraud Detection Based on Large Language Models. InIEEE Traceable LLM Reasoning for Fake-Order Fraud Detection KDD ’27, August 2027, San Jose, California International Conference on Blockchain and Trusted Intelligence Systems (ICBCTIS)

    2025

  5. [5]

    Paul F Christiano, Jan Leike, Tom Brown, Miljan Martic, Shane Legg, and Dario Amodei. 2017. Deep reinforcement learning from human preferences.Advances in neural information processing systems30 (2017)

    2017

  6. [6]

    María del Mar Roldán-García, José García-Nieto, and José F Aldana-Montes. 2017. Enhancing semantic consistency in anti-fraud rule-based expert systems.Expert Systems with Applications90 (2017), 332–343

    2017

  7. [7]

    V Dheepa and R Dhanapal. 2012. Behavior based credit card fraud detection using support vector machines.ICTACT Journal on Soft computing2, 4 (2012), 391–397

    2012

  8. [8]

    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

Show all 40 references
  1. [9]

    Kawin Ethayarajh, Winnie Xu, Niklas Muennighoff, Dan Jurafsky, and Douwe Kiela. 2024. Kto: Model alignment as prospect theoretic optimization.arXiv preprint arXiv:2402.01306(2024)

    2024 arXiv

  2. [10]

    Chenghan Fu, Daoze Zhang, Yukang Lin, Zhanheng Nie, Xiang Zhang, Jianyu Liu, Yueran Liu, Wanxian Guan, Pengjie Wang, Jian Xu, et al. 2025. MOON Embedding: Multimodal Representation Learning for E-commerce Search Advertising.arXiv preprint arXiv:2511.11305(2025)

    2025

  3. [11]

    Team GLM, Aohan Zeng, Bin Xu, Bowen Wang, Chenhui Zhang, et al . 2024. ChatGLM: A Family of Large Language Models from GLM-130B to GLM-4 All Tools. arXiv:2406.12793

    2024 arXiv

  4. [12]

    Daya Guo, Dejian Yang, Haowei Zhang, Junxiao Song, Ruoyu Zhang, Runxin Xu, Qihao Zhu, Shirong Ma, Peiyi Wang, Xiao Bi, et al . 2025. Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning.arXiv preprint arXiv:2501.12948(2025)

    2025 arXiv

  5. [13]

    Mustafa Berk Keles and Mehmet Gokturk. 2026. BI-GBDT: A Graph-Free Be- havioral Interaction-Aware Gradient Boosting Framework for Fraud Detection in Large-Scale Payment Systems.Applied Sciences16, 2 (2026). doi:10.3390/ app16020876

    2026

  6. [14]

    Zhanheng Nie, Chenghan Fu, Daoze Zhang, Junxian Wu, Wanxian Guan, Pengjie Wang, Jian Xu, and Bo Zheng. 2026. Moon2. 0: Dynamic modality-balanced multimodal representation learning for e-commerce product understanding. In Proceedings of the IEEE/CVF Conference on Computer Visio...

    2026

  7. [15]

    Long Ouyang, Jeffrey Wu, Xu Jiang, Diogo Almeida, Carroll Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, et al. 2022. Training language models to follow instructions with human feedback.Advances in neural information processing systems35 (...

    2022

  8. [16]

    Junjun Pan, Yixin Liu, Xin Zheng, Yizhen Zheng, Alan Wee-Chung Liew, Fuyi Li, and Shirui Pan. 2025. A label-free heterophily-guided approach for unsuper- vised graph fraud detection. InProceedings of the AAAI Conference on Artificial Intelligence, Vol. 39. 12443–12451

    2025

  9. [17]

    Tahereh Pourhabibi, Kok-Leong Ong, Booi H Kam, and Yee Ling Boo. 2020. Fraud detection: A systematic literature review of graph-based anomaly detection approaches.Decision Support Systems133 (2020), 113303

    2020

  10. [18]

    Rafael Rafailov, Archit Sharma, Eric Mitchell, Christopher D Manning, Stefano Ermon, and Chelsea Finn. 2023. Direct preference optimization: Your language model is secretly a reward model.Advances in neural information processing systems36 (2023), 53728–53741

    2023

  11. [19]

    Nils Reimers and Iryna Gurevych. 2019. Sentence-bert: Sentence embeddings using siamese bert-networks.arXiv preprint arXiv:1908.10084(2019)

    2019 arXiv

  12. [20]

    John Schulman, Filip Wolski, Prafulla Dhariwal, Alec Radford, and Oleg Klimov

  13. [21]

    2017.Credit scoring and its applications

    Lyn Thomas, Jonathan Crook, and David Edelman. 2017.Credit scoring and its applications. SIAM

    2017

  14. [22]

    Che-Ping Tsai, Ganyu Teng, Phillip Wallis, and Wei Ding. 2025. AnoLLM: Large Language Models for Tabular Anomaly Detection. InInternational Conference on Learning Representations (ICLR)

    2025

  15. [23]

    Ruiqi Wang, Ruikang Liu, Runyu Chen, Haoxiang Suo, Zhiyi Peng, Zhuo Tang, and Changjian Chen. 2026. CausalTAD: Injecting Causal Knowledge into Large Language Models for Tabular Anomaly Detection.arXiv preprint arXiv:2602.07798 (2026)

    2026

  16. [24]

    Ronald J Williams. 1992. Simple statistical gradient-following algorithms for connectionist reinforcement learning.Machine learning8, 3 (1992), 229–256

    1992

  17. [25]

    Junxian Wu, Chenghan Fu, Zhanheng Nie, Daoze Zhang, Bowen Wan, Wanxian Guan, Chuan Yu, Jian Xu, and Bo Zheng. 2026. MOON3. 0: Reasoning-aware Multimodal Representation Learning for E-commerce Product Understanding. arXiv preprint arXiv:2604.00513(2026)

    2026

  18. [26]

    Jianhao Yan, Yafu Li, Zican Hu, Zhi Wang, Ganqu Cui, Xiaoye Qu, Yu Cheng, and Yue Zhang. 2025. Learning to Reason under Off-Policy Guidance. arXiv:2504.14945 [cs.LG] https://arxiv.org/abs/2504.14945

    2025 arXiv

  19. [27]

    An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, et al. 2025. Qwen3 Technical Report. arXiv:2505.09388 [cs.CL] https://arxiv.org/abs/2505.09388

    2025 arXiv

  20. [28]

    Daoze Zhang, Chenghan Fu, Zhanheng Nie, Jianyu Liu, Wanxian Guan, Yuan Gao, Jun Song, Pengjie Wang, Jian Xu, and Bo Zheng. 2026. MOON: Generative MLLM- based multimodal representation learning for e-commerce product understanding. InProceedings of the Nineteenth ACM Internatio...

    2026

  21. [29]

    Yuze Zhao, Jintao Huang, Jinghan Hu, Xingjun Wang, Yunlin Mao, et al

  22. [33]

    Second point of analysis... </reason> The most likely motive scenario is: Scenario name Confidence level: [High/Medium/Low] Additional notes: [If any] ## Basic transaction information of the order {{feature}} ## Manual analysis {{manual_analysis}} ## Transaction motive {{true_...

    2027

  23. [35]

    Second point of analysis... </reason> The most likely motive scenario is: Scenario name Confidence level: [High/Medium/Low] Additional notes: [If any] ## Basic transaction information of the order {{feature}} ## Transaction motive {{true_label}} Traceable LLM Reasoning for Fak...

    2027

  24. [37]

    Second point of analysis... </reason> <answer> The most likely motive scenario is: Scenario name Confidence level: [High/Medium/Low] Additional notes: [If any] </answer> ## Basic transaction information of the order {{feature}} Prompt of Self-verification ## Task Description Y...

  25. [39]

    Second point of analysis... </reason> <answer> The most likely motive scenario is: Scenario name Confidence level: [High/Medium/Low] Additional notes: [If any] </answer> KDD ’27, August 2027, San Jose, California You et al. Prompt of Reflection ## Task Description Please gradu...

    2027

  26. [40]

    First point of analysis

  27. [41]

    Second point of analysis... </reason> <answer> The most likely motive scenario is: Scenario name Confidence level: [High/Medium/Low] Additional notes: [If any] </answer> Prompt of Suggestions from Transaction-expert ## Role and Task As an expert in O2O platform business, pleas...

    2027

  28. [42]

    State the analysis logic step by step

  29. [43]

    Only speak based on the provided feature data

  30. [44]

    </reason> <answer> There is an abnormal transaction./There is not an abnormal transaction

    Mark key judgment basis. </reason> <answer> There is an abnormal transaction./There is not an abnormal transaction. </answer>

  31. [2017]

    Proximal policy optimization algorithms.arXiv preprint arXiv:1707.06347 (2017)

    2017 arXiv

  32. [2024]

    A de-identified entity group contains 17 linked accounts, with aggregated activity observed over the recent pe- riod

    SWIFT:A Scalable lightWeight Infrastructure for Fine-Tuning. arXiv:2408.05517 [cs.CL] https://arxiv.org/abs/2408.05517 KDD ’27, August 2027, San Jose, California You et al. A Appendix A.1 Ethics and Privacy This study strictly adheres to ethical standards and robust data secur...

    2027 arXiv

Pith tools

Reviewed August 1, 2026 · model on record in the stance chip above.