G-SHARE, a new guideline-based structured reasoning framework, aims to improve human-factor event diagnosis in nuclear power plants. Developed by a team of researchers including Xingyu Xiao, Mao Du, Jiejuan Tong, Jingang Liang, and Haitao Wang, this framework was introduced in a paper submitted on May 7, 2026.
The quality of human-factor event diagnosis relies heavily on expert interpretation of narrative reports. However, traditional data-driven or one-shot large language model approaches often lack structured reasoning, which can lead to inconsistencies in conclusions. G-SHARE addresses this challenge by operationalizing the CNNP nine-step human-factor event diagnosis guideline into a comprehensive multi-stage diagnostic framework.
Key Features of G-SHARE Framework
The G-SHARE framework consists of three main components:
- Evidence Extraction: Systematically gathers relevant data from reports.
- Stepwise Diagnostic Reasoning: Follows a structured approach to analyze and interpret findings.
- Post-Hoc Consistency Repair: Validates outcomes to ensure logical coherence.
This structured reasoning approach allows for explicit use of report evidence, intermediate rationale generation, and logical validation of diagnostic outputs, making it a valuable tool for safety-critical industries.
Performance and Evaluation
A dataset of real human-factor event reports was constructed from sources within the Chinese nuclear industry. A gold-standard subset was annotated by domain experts for evaluation purposes. Results indicate that G-SHARE significantly outperforms traditional machine learning baselines and one-shot prompting techniques. The strongest version of G-SHARE achieved the best overall accuracy and macro-F1 scores.
Ablation studies revealed that both structured reasoning and consistency enforcement are critical for robust diagnosis, particularly under weak prompting conditions. These findings underscore the importance of transforming expert diagnostic guidelines into auditable reasoning workflows.
Implications for Safety-Critical Industries
The introduction of G-SHARE presents a practical pathway for intelligent human-factor analysis, particularly in safety-critical environments like nuclear power. By enhancing the diagnostic process, this framework not only improves the reliability of event diagnoses but also contributes to overall operational safety.
In conclusion, G-SHARE represents a significant advancement in structured reasoning frameworks, demonstrating the potential to improve human-factor event diagnosis through rigorous adherence to established guidelines.
🤖 This article was rewritten by Feed and Figures' editorial AI from a report originally published by arXiv NLP. Facts and quotes are preserved from the original; the rewrite focuses on clarity and structure. For the unedited original, see the source link below.
