What Are the Facts? Automated Extraction of Court-Established Facts from Criminal-Court Opinions

Overview

The paper explores the feasibility of using automated methods to extract detailed descriptions of criminal behaviors from Slovak court decisions. This is achieved through advanced regular expressions and large language models (LLMs), which significantly outperform baseline methods.

Key Insights

• Extraction Accuracy: Advanced regular expressions achieved 97% accuracy, LLMs 98.75%, and a combination of both reached 99.5%.
  • Evidence: Evaluation by law students showed a 90% match with human annotations.
  • Verifiable: Yes, through replication of the study.
• Baseline Comparison: The baseline method identified descriptions in only 40.5% of cases.
  • Evidence: Compared to 34.5% match with human annotations.
  • Verifiable: Yes, through study data.

BFSI Relevance

• Why Relevant: Enhanced data extraction methods can improve the accuracy and efficiency of legal data analysis, impacting risk assessment and compliance in financial services.
• Primary Sector: Financial Services
• Subsectors: Legal Compliance, Risk Management
• Actionable Implications:
  • Implement similar automated extraction methods to improve data processing.
  • Use enhanced data insights for better compliance and risk management strategies.