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

Abstract

Court proceedings contain valuable evidence about human smuggling networks, but this information is often buried within unstructured, jargon-heavy legal documents. While large language models (LLMs) can support knowledge graph construction through automated information extraction, existing approaches rely on general-purpose models that are not tailored to the entity and relationship definitions required in this domain. We introduce FineREX, a streamlined knowledge graph construction pipeline built around a fine-tuned LLM for named entity recognition and relationship extraction (NER-RE). Using a manually annotated dataset of text chunks, FineREX achieves absolute improvements of 15.50% and 31.46% in entity and relationship F1-score, respectively, compared to a larger general-purpose baseline. These gains translate into higher-quality knowledge graphs, reducing legal noise by nearly half and lowering node duplication on long documents from 17.78% to 11.17%. By eliminating document rewriting and redundant extraction stages, FineREX also reduces end-to-end processing time by 50.0%. Our results demonstrate that domain-specific fine-tuning can substantially outperform larger general-purpose models while improving both the quality and efficiency of knowledge graph construction for illicit network analysis.

FineREX is a framework for domain-specific knowledge graph construction, specializing in human smuggling. It introduces a fine-tuned approach for Named Entity Recognition and Relationship Extraction (NER-RE) using Llama 3.1 models, outperforming generic baseline pipelines.

Our fine-tuned model may be accessed at Model, and run with ollama run 2028efeldman/llama-finetuned.

Project Structure

FineREX (Core Implementation)

The scripts/ directory contains the implementation of the FineREX approach:

• scripts/llama_finetune/: Core training and inference logic.
  • train.py: QLoRA-based fine-tuning for NER-RE.
  • run_splits.py: Batch inference using fine-tuned models.
• scripts/kg/: Knowledge Graph construction and consolidation.
  • build_consolidated_kg_networkx.py: The primary FineREX script for merging model extractions into canonicalized KGs using NetworkX.
  • build_eval_case_kgs.py: Generates KGs for specific evaluation cases.
• scripts/util/: Utilities for bridging model outputs to graph processing.

LinkKG (Baseline)

The linkkg/ directory contains the baseline pipeline used for comparison:

• linkkg/run_pipeline.py: A modular pipeline stage manager.
• linkkg/ner.py, linkkg/loopcoref.py, linkkg/resolve_coref.py: Baseline stages using non-specialized or 70B models.
• linkkg/generate_kgs.py: Simple KG generation for baseline results.

Data and Guidelines

• dataset/guidelines.md: Comprehensive annotation guidelines and entity definitions (PERSON, LOCATION, ORGANIZATION, etc.).
• setup/requirements.txt: Environment dependencies.

Installation

Prerequisites

• Python 3.12
• CUDA-compatible GPU

Setup

We recommend using uv for environment management:

uv venv --python 3.12
source .venv/bin/activate
pip install -r setup/requirements.txt

FineREX Pipeline

Fine-tuning Process

Configure scripts/llama_finetune/config.py with your dataset paths and base model, then run:

python -m scripts.llama_finetune.train

For evaluation and NER-RE extraction of this model, use:

python -m scripts.llama_finetune.run_llama8b_splits --output-root runs/finetune_results

Coreference Resolution

bash scripts/util/run_coref_finetune.sh

KG Consolidation

Construct the final canonicalized Knowledge Graph from the model's metrics/predictions:

python scripts/kg/build_consolidated_kg_networkx.py

Baseline Comparison (LinkKG)

To run the baseline pipeline for comparison:

python linkkg/run_pipeline.py \
    --input-file-name "case_name" \
    --entity-type "PERSON" \
    --run-stages prep chunk ner coref resolve \
    --ner-model-name "llama3.1:70b" \
    --ner-prompt-file "linkkg/prompts/person_nopr_ner_prompt.txt"

Citation

@article{feldmanfinerex2026,
  title   = {{FineREX}: Fine-Tuned {NER-RE} for Human Smuggling Knowledge Graphs},
  author  = {Feldman, Elijah and Meher, Dipak and Domeniconi, Carlotta},
  journal = {arXiv preprint arXiv:2606.19710},
  year    = {2026},
  url     = {[https://arxiv.org/abs/2606.19710](https://arxiv.org/abs/2606.19710)}
}