@article{ART003338880},
author={Seungkeon Lee and Namgyu Kim},
title={Window-Augmentation Based Methodology for Improving Error Attribution Performance in Multi-Agent Systems},
journal={Journal of The Korea Society of Computer and Information},
issn={1598-849X},
year={2026},
volume={31},
number={5},
pages={55-66}
TY - JOUR
AU - Seungkeon Lee
AU - Namgyu Kim
TI - Window-Augmentation Based Methodology for Improving Error Attribution Performance in Multi-Agent Systems
JO - Journal of The Korea Society of Computer and Information
PY - 2026
VL - 31
IS - 5
PB - The Korean Society Of Computer And Information
SP - 55
EP - 66
SN - 1598-849X
AB - Recent LLM Multi-Agent Systems have demonstrated strong performance across various domains; however, they are limited in that errors can propagate across steps due to their interaction structure.
Accordingly, the importance of failure attribution, which aims to identify the root cause of task failure, has been increasingly emphasized. To address the limitation of existing methods that do not sufficiently capture interactions between adjacent steps, this study proposes a window-augmentation-based methodology. The proposed approach constructs a localized context around error steps identified by existing methods and re-identifies the decisive error step and responsible agent. Experimental results show that the proposed method improves both step and agent accuracy over existing approaches.
Notably, the best performance is achieved with a window size of 3, highlighting the importance of proper window size selection. These findings demonstrate that localized context-based analysis is effective for failure attribution in multi-agent systems and can be extended to diverse environments.
KW - LLM Multi-Agent Systems;Failure Attribution;Window Augmentation;Local Context
DO -
UR -
ER -
Seungkeon Lee and Namgyu Kim. (2026). Window-Augmentation Based Methodology for Improving Error Attribution Performance in Multi-Agent Systems. Journal of The Korea Society of Computer and Information, 31(5), 55-66.
Seungkeon Lee and Namgyu Kim. 2026, "Window-Augmentation Based Methodology for Improving Error Attribution Performance in Multi-Agent Systems", Journal of The Korea Society of Computer and Information, vol.31, no.5 pp.55-66.
Seungkeon Lee, Namgyu Kim "Window-Augmentation Based Methodology for Improving Error Attribution Performance in Multi-Agent Systems" Journal of The Korea Society of Computer and Information 31.5 pp.55-66 (2026) : 55.
Seungkeon Lee, Namgyu Kim. Window-Augmentation Based Methodology for Improving Error Attribution Performance in Multi-Agent Systems. 2026; 31(5), 55-66.
Seungkeon Lee and Namgyu Kim. "Window-Augmentation Based Methodology for Improving Error Attribution Performance in Multi-Agent Systems" Journal of The Korea Society of Computer and Information 31, no.5 (2026) : 55-66.
Seungkeon Lee; Namgyu Kim. Window-Augmentation Based Methodology for Improving Error Attribution Performance in Multi-Agent Systems. Journal of The Korea Society of Computer and Information, 31(5), 55-66.
Seungkeon Lee; Namgyu Kim. Window-Augmentation Based Methodology for Improving Error Attribution Performance in Multi-Agent Systems. Journal of The Korea Society of Computer and Information. 2026; 31(5) 55-66.
Seungkeon Lee, Namgyu Kim. Window-Augmentation Based Methodology for Improving Error Attribution Performance in Multi-Agent Systems. 2026; 31(5), 55-66.
Seungkeon Lee and Namgyu Kim. "Window-Augmentation Based Methodology for Improving Error Attribution Performance in Multi-Agent Systems" Journal of The Korea Society of Computer and Information 31, no.5 (2026) : 55-66.
