cs.CLFeb 27, 2026

Reviewing the Reviewer: Graph-Enhanced LLMs for E-commerce Appeal Adjudication

Authors: Yuchen Du, Ashley Li, Zixi Huang

Organizations: Purdue University West Lafayette, Indiana, USA · Amazon Seattle, Washington, USA

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

Hierarchical review workflows, where a second-tier reviewer (Checker) corrects first-tier (Maker) decisions, generate valuable correction signals that encode why initial judgments failed. However, learning from these signals is hindered by information asymmetry: corrections often depend on verification actions unavailable to Makers or automated systems. We address this challenge by introducing explicit action modeling as an inferential constraint that grounds reasoning in verifiable operations rather than unconstrained text generation. We propose the Evidence-Action-Factor-Decision (EAFD) schema, a minimal representation for adjudication reasoning that prevents hallucination through operational grounding and enables learning from correction signals via explicit conflict modeling. Building on this schema, we develop a conflict-aware graph reasoning framework that: (1) constructs EAFD graphs from historical cases capturing Maker-Checker disagreements, (2) aggregates them into a retrievable knowledge base, and (3) performs top-down deductive reasoning for new cases by projecting validated resolution paths from precedents. A distinctive capability is the Request More Information (RMI) outcome: when evidence is insufficient, the system identifies precisely which verification actions remain unexecuted and generates targeted information requests. We evaluate the framework in large-scale e-commerce seller appeal adjudication. While a standard LLM-only baseline achieves only 70.8% alignment with human experts, incorporating action modeling with RMI improves alignment to 87.5%. Augmenting this with the retrieval-based knowledge graph yields the best offline performance of 95.8%. Following online deployment, the framework maintains robust performance, achieving a 96.3% alignment rate in production, demonstrating its real-world effectiveness.

Figures & tables

Figure 1. System overview. Offline: Extract EAFD graphs from historical cases and build a knowledge base. Online: Construct Maker graph, retrieve precedents, derive Checker graph via FAE deduction, and output adjudication decision.
Figure 2. The EAFD schema with four-layer hierarchy. Left: Maker’s initial review leads to rejection. Right: Checker’s corrective review with additional verification overturns the decision. The conflict edge (red) captures Maker-Checker disagreement as a learning signal.
Figure 3. Online reasoning pipeline. The system constructs the Maker graph, retrieves similar cases, aligns factors to historical paths, and applies FAE deduction ( fq(C)→aq(C)→eq(C)f_{q}^{(C)}\to a_{q}^{(C)}\to e_{q}^{(C)} ). Final decision d^q(C)\hat{d}_{q}^{(C)} is based on action status; RMI is triggered when critical evidence is missing.
MethodUsesLLMActionRMI
 HistoryReasoningGroundingEnabled
LLM (Query-Only)✗✓✗✗
LLM + RMI✗✓✓✓
CBR (Majority Vote)✓✗✗✗
LLM + CBR✓✓✗✗
LLM + CBR + RMI✓✓✓✓
Table 1. Baseline taxonomy along four design dimensions.
 Overall Metrics
MethodAcc.Macro F1Macro Recall
LLM (Query-Only)0.7080.6040.728
LLM + RMI0.8750.6820.760
CBR (Majority Vote)0.3700.2500.322
LLM + CBR0.6300.3420.300
LLM + CBR + RMI0.6670.3540.342
Table 2. Overall adjudication performance under different reasoning assumptions.
Figure 4. Cumulative alignment rate in production (Jan 17 – Feb 5).
ClassPrecisionRecallF1# Case
Approve (Reinstate)1.0000.9790.98948
Reject0.9880.9660.97787
Request More Info0.2501.0000.4001
Accuracy97.1% (132 / 136)
Table 3. Online per-class performance over the first 136 cases.
Appendix figures & tables1 asset

Supplementary material from the paper’s appendix.

Appendix

 Overturn CasesNon-Overturn Cases
Avg. Actions (System)4.92.0
Action Hit Rate0.810.74
Cases with ≥\geq 1 Missing Action18%41%
Appendix
Table 4. Action-level analysis comparing overturn and non-overturn appeal cases. “Action Hit Rate” measures the proportion of system-recommended verification actions that semantically match at least one Checker-documented action in the final review record.
More in this topic

LLM Decision-Making

LLM decision-making studies how language models choose, act, seek information, and allocate resources under goals, uncertainty, constraints, and social pressures.

See all 242 papers

Explore similar work

CardsList
Sep 22, 2026

JEV-as-a-Judge: Accept When Confident, Escalate When Unsure

LLM-as-a-judge scales evaluation, but reasoning judges are slow and costly. We study JEV-as-a-Judge: evaluation with JEV, a decision-only judge that returns label probabilities instead of text, and whose confidence decides whether to accept its verdict or escalate to a reasoning judge. Against sixteen generative and reward-model judges, with blinded human adjudication, JEV comes within three points of GPT-6 wherever a verdict can be read off the text, at 0.36% of its fee and a 0.15-second median latency, and falls behind where the verdict must be derived, as in math, code, and logic. Its confidence marks this boundary. With a threshold frozen in advance, accepting confident verdicts and escalating the rest is 0.9 points more accurate than GPT-6 on 1,610 held-out pairs at 41% of its fee, and in a pre-specified live test on two new workloads the cascade matches GPT-6's accuracy exactly. Confidence routing weakens on style-adversarial pairs and reference-free prose; we close with a simple recipe for validating thresholds locally.

Yubo Li, Yidi Miao, Ramayya Krishnan +1

Carnegie Mellon University

LLM-as-a-JudgeLanguage Model Generation EvaluationSelective PredictionLLM Uncertainty EstimationAutomated Evaluation
AskPDF
May 27, 2026

CyberJurors: A Multi-Agent Simulation Task for E-Commerce Disputes Verdict

E-commerce platforms have begun recruiting crowdsourced jurors to adjudicate massive volumes of transaction disputes. Unlike formal legal judgment, E-commerce dispute verdicts require grounding pivotal clues from redundant, multi-round, multimodal evidence and making decisions under flexible platform-specific conventions. These characteristics render existing methods insufficient for this scenario. To bridge this gap, we introduce a pioneering task, E-commerce Dispute Verdicts (EDV), and present VerdictBench, a multimodal benchmark comprising 6,000 real-world cases designed to reflect crowdsourced jury decisions. Building upon this, we propose CyberJurors, a multi-agent framework to clarify the dispute logic and regulate the verdict process. At the individual level, Individual Verdict Chain-of-Thought decomposes the EDV task into four structured reasoning stages, enabling fine-grained clue perception and clarifying causal logic between pivotal clues and the dispute focus. At the collective level, Jury Consensus Verdict simulates multi-round discussion and voting among jurors, while incorporating verdict precedents to mitigate cognitive biases toward either disputant. Experiments on VerdictBench show that CyberJurors outperforms state-of-the-art LLMs, MLLMs, and court simulators, while achieving stronger alignment with real-world jury voting patterns. Code and dataset are available at https://github.com/YanhuiS/CyberJurors and https://huggingface.co/datasets/piggi/VerdictBench.

Yanhui Sun, Wu Liu, Haifeng Ming +3

School of Information Science and Technology, University of Science and Technology of China, Hefei, China.

Multimodal ReasoningLLM Decision-MakingE-CommerceAgent Benchmarks
AskPDF
Jun 16, 2026

EComAgentBench: Benchmarking Shopping Agents on Long-Horizon Tasks with Distributed Hidden Intent

As LLM-based shopping agents enter production, existing benchmarks fail to capture how a shopper's requirements arrive: stated implicitly in the query, recorded in a profile, or revealed only when the right question is asked. Benchmarks that expose full intent upfront and grade only the final choice can neither pose this long-horizon challenge nor explain which requirement an agent missed. To address this gap, we introduce EComAgentBench, a benchmark of 662 tasks grounded in real Amazon products and reviews. Each task scatters these requirements across a visible query, a tool-gated profile, and scripted clarification; an agent must uncover hidden intent, verify candidates against attributes and review evidence, and commit to a single product within 100 tool calls. Moreover, typed, source-tagged rubrics grade every task, attributing each failure to a requirement and its source. Construction is automated yet reliable, with every answer fixed in code before any text is generated and every sample validated. Our evaluation of seven models reveals that even the strongest attains only 57.1% overall accuracy, and rubric satisfaction degrades from visible to hidden sources. Overall, we believe EComAgentBench will serve as a reproducible foundation for moving shopping agents from single-query search toward dependable assistance over long horizons.

Zeyao Du, Tong Li, Haibo Zhang

Shopee

Web AgentsLong-Horizon Agent EvaluationAI Agent BenchmarksTool-Use EvaluationAgentic CommercePreference Elicitation+1
AskPDF