Criminal Justice Entity Classifier

This solution tags entities in criminal justice documents as either government or person using finite rules, then uses large language models to create custom training data with those labels. The approach aims to improve the accuracy of entity classification in legal and criminal justice contexts. The system was developed for Georgia State University as part of the SCALES challenge at the KnowHax Hackathon Episode 1 in May 2024.

The team combined rule-based classification with LLM-generated training data. They first applied finite rules to identify and tag entities as government or person, then used those labeled examples to prompt large language models to generate additional custom training data for improving the classification model.

Georgia State University benefits from this solution, which addresses the challenge of accurately classifying entities in criminal justice documents.