Software

conflibertR, the NL Agent Explorer, ConfliBERT GUI and open-weight language models for conflict text.

Tools

conflibertR hex logo

conflibertR

An R package bringing ConfliBERT, the NSF-funded language model for political conflict and violence, to the R ecosystem: entity recognition, event classification, question answering, fine-tuning, model comparison, and active learning, all from R. On CRAN since July 2026.

NL Agent Explorer

Interactive explorer of agent-level data on Dutch national politics, 2008 to 2025: every Tweede Kamer member and cabinet minister with offices held, party and parliamentary group, background, committee seats, bills sponsored with their Tweede Kamer topics and Raad van State advice, and a yearly stance on executive power estimated from plenary speech. Built for the TWIN4DEM project.

ConfliBERT GUI

Browser-based NLP toolkit for conflict and political violence text analysis: named entity recognition, classification, question answering, and LoRA/QLoRA fine-tuning.

Fine-tuned language models

Open-weight models for conflict and political text, all on Hugging Face.

ConflLlama

Llama 3 8B fine-tuned on the Global Terrorism Database for attack-type classification. Also available with a more neutral chat template and in four time-period variants.

ConflLlama-NER

Llama 3.1 8B Instruct fine-tuned for named entity recognition in conflict event narratives: the source, target, and related actors of each event.

ConflLlama-NER-TOI

Named entity recognition for Indian news (locations, organisations, people, and time expressions), trained on 100,000 Times of India articles.

Qwen-GLOCON-Reasoning

Qwen2.5-3B trained with GRPO reinforcement learning to classify conflict events into five categories with structured reasoning.

Confli-mBERT

ModernBERT fine-tuned on the Global Terrorism Database for conflict event classification; companion model to the paper on NLP model selection in computational social science.