Hybrid Framework for Causality Extraction in Nuclear Licensee Event Reports

Abstract

The accurate extraction of causality in nuclear licensee event reports is critical for enhancing safety measures and operational efficiency within the nuclear industry. This study presents a hybrid framework designed to improve the precision and reliability of causality extraction from these reports. The proposed framework integrates rule-based methods with machine learning techniques to leverage the strengths of both approaches. A comprehensive dataset of nuclear event reports was utilized to train and evaluate the system, demonstrating significant improvements in causality detection accuracy compared to existing methodologies. The framework's robustness is further validated through cross-validation and comparison with human expert analyses. The results indicate that the hybrid approach not only enhances the granularity and context-awareness of extracted causal relationships but also reduces the time and effort required for manual report analysis. This advancement holds promise for more effective risk assessment and management in nuclear operations.