Using text mining and multilevel association rules to process and analyze incident reports in China

Area of Science:

• Safety Science
• Data Mining
• Natural Language Processing

Background:

• Current analysis of incident reports relies on expert experience, leading to under-analysis of minor incidents and near misses.
• This reliance on expertise and omission of minor incidents limits comprehensive safety improvement strategies.
• A need exists for efficient, data-driven methods to analyze incident reports and extract actionable safety insights.

Purpose of the Study:

• To propose a framework using text mining and multilevel association rules for structuring Chinese incident reports.
• To identify critical incident patterns, analyze trends, and guide safety management strategies.
• To demonstrate the framework's utility through a case study in the Chinese construction industry.

Main Methods:

• Developed a pattern extraction workflow using TextRank and domain pertinence for Chinese reports.
• Applied a concept hierarchy to establish taxonomic relationships among risk factors.
• Utilized multilevel association rule mining for comprehensive pattern identification and analysis.

Main Results:

• The framework efficiently structured incident data and identified significant incident patterns.
• Analysis revealed temporal features of incidents and the effectiveness of safety measures.
• Near-miss events were highlighted as crucial for learning and improving safety performance.

Conclusions:

• The proposed framework offers an efficient method for analyzing Chinese incident reports, moving beyond expert reliance.
• Multilevel association rules provide deeper insights into incident patterns and safety trends.
• Leveraging near-miss data is vital for proactive safety management and policy formulation.