SIRIS-Lab / org_type_classifier-v2

🌍 Organization Type Classifier (v2)

This is a fine-tuned Transformer model for multilingual classification of organization types , developed by SIRIS Lab . It takes as input the name of an organization and predicts the corresponding type.

🧠 Model Overview

• Base model : intfloat/multilingual-e5-large , initialized from xlm-roberta-large
• Number of classes : 6
• Input : Only the organization name , no description
• Hint : Capitalizing the organization name often improves classification performance

🏷️ Labels

The classifier predicts one of the following organizational types:

🔍 Intended Use

This model is designed for automatic classification of organisations found in datasets, publications, or project records — such as for:

• Metadata enrichment
• Affiliation normalization
• Analytical typologies in research and innovation studies

It performs well on multilingual inputs , especially in European languages.

📥 Input Format

A single organization name (ideally capitalized), e.g.:

UNIVERSITA DEGLI STUDI DI ROMA TOR VERGATA

📤 Output Format

A predicted label from the list above.

💾 Files

The repository includes:

• config.json : Model configuration
• model.safetensors : Model weights
• tokenizer.json , tokenizer_config.json , sentencepiece.bpe.model , special_tokens_map.json : Tokenizer files
• training_args.bin : HF training arguments
• test_predictions.csv : Predictions on a held-out test set

📊 Performance

The model performs well on a manually validated multilingual dataset of organization names. It is especially accurate for well-known institutions and clearly named entities.

Note: Borderline or ambiguous names (e.g. acronyms, hybrids) may need contextual post-processing.

🛠 How to Use

from transformers import pipeline

classifier = pipeline("text-classification", model="SIRIS-Lab/org_type_classifier-v2", tokenizer="SIRIS-Lab/org_type_classifier-v2")
classifier("UNIVERSITY OF OXFORD")
# Output: [{'label': 'TED', 'score': ...}]

📚 Citation

If you use this model, please cite or acknowledge SIRIS Lab .

🔒 License

MIT