Universal Named Entity Recognition (NER) with GLiNER

Short summary

This repo is intended to share the Congruence Engine’s experiments with fine-tuning a version of the GLiNER (Generalist and Lightweight Model for Named Entity Recognition) model. This model is part of a new wave of NER models commonly referred to as ‘Universal NER’ – the key distinction from traditional NER being that the model is not restricted to previously established entity types, but can entities based on user-defined labels.

Research question(s):

• How useful are Universal NER models in the context of museums cultural heritage?
• What kinds of linkage do these new models lead to?

People

Max Long

Investigation, Data curation, Formal analysis, Methodology, Writing

Arran Rees

Investigation, Data curation, Formal analysis, Methodology, Writing

Kaspar Beelen

Methodology, Software

Data sources

• GLiNER machine learning NER model (Huggingface)
• Textile machinery datasets from Science Museum, National Wool Museum, and Bradford Industrial Museum
• Various textile glossaries sourced from archive.org.

Investigation methods/ tools/ code/ software

• Named Entity Recognition
• Text Classification
• Large Language Models including Chat GPT

Outputs

• Fine-tuned GLiNER models
• Hugginface demo of the fine-tuned model in action
• Google Colab notebook tutorial on how to fine-tune the model in the context of cultural heritage (to be completed)

Licence

This work is licensed under a Creative Commons Attribution 4.0 License - CC BY 4.0.