@article{ART003370680},
author={Sharon Kim and Seok-Yong Yun},
title={Numerical Noise Detection in E-commerce Product Data Using a Category-Relative Ordinal AUM-Based LLM Pipeline},
journal={Journal of The Korea Society of Computer and Information},
issn={1598-849X},
year={2026},
volume={31},
number={8},
pages={41-49}
TY - JOUR
AU - Sharon Kim
AU - Seok-Yong Yun
TI - Numerical Noise Detection in E-commerce Product Data Using a Category-Relative Ordinal AUM-Based LLM Pipeline
JO - Journal of The Korea Society of Computer and Information
PY - 2026
VL - 31
IS - 8
PB - The Korean Society Of Computer And Information
SP - 41
EP - 49
SN - 1598-849X
AB - E-commerce catalogs contain numerical noise that stays within the statistical range of a category and thus evades distribution-based detection. This study proposes REPAIR, a training-free LLM pipeline that detects semantic inconsistency between product names and weight values by mapping weights into category-relative ordinal bins and scoring the margin and ordinal distance between the bin predicted via product-contextual Chain-of-Thought reasoning and the observed bin (Category-Relative Ordinal AUM).
On 798 Amazon Berkeley Objects samples with anomalies constructed by In-Distribution Semantic Swap over five seeds, REPAIR records an AUROC of 0.844±0.012 (+0.128 over Standard AUM) and 0.824– 0.868 across three LLMs, confirming model-agnostic behavior.
KW - Anomaly Detection;Large Language Model;E-commerce;Area Under the Margin;Chain-of-Thought;Data Quality
DO -
UR -
ER -
Sharon Kim and Seok-Yong Yun. (2026). Numerical Noise Detection in E-commerce Product Data Using a Category-Relative Ordinal AUM-Based LLM Pipeline. Journal of The Korea Society of Computer and Information, 31(8), 41-49.
Sharon Kim and Seok-Yong Yun. 2026, "Numerical Noise Detection in E-commerce Product Data Using a Category-Relative Ordinal AUM-Based LLM Pipeline", Journal of The Korea Society of Computer and Information, vol.31, no.8 pp.41-49.
Sharon Kim, Seok-Yong Yun "Numerical Noise Detection in E-commerce Product Data Using a Category-Relative Ordinal AUM-Based LLM Pipeline" Journal of The Korea Society of Computer and Information 31.8 pp.41-49 (2026) : 41.
Sharon Kim, Seok-Yong Yun. Numerical Noise Detection in E-commerce Product Data Using a Category-Relative Ordinal AUM-Based LLM Pipeline. 2026; 31(8), 41-49.
Sharon Kim and Seok-Yong Yun. "Numerical Noise Detection in E-commerce Product Data Using a Category-Relative Ordinal AUM-Based LLM Pipeline" Journal of The Korea Society of Computer and Information 31, no.8 (2026) : 41-49.
Sharon Kim; Seok-Yong Yun. Numerical Noise Detection in E-commerce Product Data Using a Category-Relative Ordinal AUM-Based LLM Pipeline. Journal of The Korea Society of Computer and Information, 31(8), 41-49.
Sharon Kim; Seok-Yong Yun. Numerical Noise Detection in E-commerce Product Data Using a Category-Relative Ordinal AUM-Based LLM Pipeline. Journal of The Korea Society of Computer and Information. 2026; 31(8) 41-49.
Sharon Kim, Seok-Yong Yun. Numerical Noise Detection in E-commerce Product Data Using a Category-Relative Ordinal AUM-Based LLM Pipeline. 2026; 31(8), 41-49.
Sharon Kim and Seok-Yong Yun. "Numerical Noise Detection in E-commerce Product Data Using a Category-Relative Ordinal AUM-Based LLM Pipeline" Journal of The Korea Society of Computer and Information 31, no.8 (2026) : 41-49.
