Automated Entity Extraction for Building Crime Anatomy from Preliminary Investigation Reports

The increasing volume and complexity of legal documents in criminal investigations pose significant challenges for investigators, particularly due to reliance on manual document review processes that are time-consuming and prone to human error. This study aims to develop an automated system for extracting and structuring key information from preliminary investigation reports to support efficient case analysis. The research employs a Natural Language Processing (NLP) approach using Named Entity Recognition (NER) based on transformer models such as BERT, combined with rule-based Regular Expressions (Regex) to construct a structured “crime anatomy.” The methodology involves data collection from Indonesian police reports, manual annotation using the IOB tagging scheme, model training and evaluation using precision, recall, and F1-score, and implementation through a multi-stage processing pipeline. The findings indicate that the integration of NER and Regex significantly improves entity extraction accuracy and structural consistency, with hybrid models achieving near-perfect performance compared to standalone NER models. The system successfully transforms unstructured legal narratives into structured summaries containing key elements such as actors, time, location, modus operandi, and evidence. In conclusion, the proposed approach enhances the efficiency, accuracy, and reliability of criminal case analysis, offering a scalable solution for modern law enforcement and contributing to improved decision-making processes.