FineREX: Fine-Tuned NER-RE for Human Smuggling Knowledge Graphs

FineREX is a fine-tuned LLM designed for named entity recognition and relationship extraction (NER-RE) specifically for constructing knowledge graphs from legal documents related to human smuggling. It utilizes a manually annotated dataset of 512 text chunks, achieving a 15.50% improvement in entity F1-score and a 31.46% improvement in relationship F1-score compared to a general-purpose model, while also reducing legal noise and node duplication. This approach highlights the effectiveness of domain-specific fine-tuning in enhancing both the quality and efficiency of knowledge graph construction, which is crucial for practitioners analyzing illicit networks.