| Property | Value |
|---|---|
| Parameter Count | 177M |
| License | CC BY-NC-SA 4.0 |
| Languages Supported | German, English, Spanish, French, Italian, Dutch, Polish, Portuguese, Russian |
| Downloads | 361,305 |
| Framework | PyTorch |
What is wikineural-multilingual-ner?
WikiNeural Multilingual NER is a state-of-the-art named entity recognition model that supports 9 different languages. Built by Babelscape, it's based on mBERT and fine-tuned on the WikiNEuRal dataset, specifically designed to address the challenge of data scarcity in multilingual NER tasks.
Implementation Details
The model utilizes a multilingual BERT architecture fine-tuned for 3 epochs on the WikiNEuRal dataset. It combines both neural and knowledge-based approaches for silver data creation, resulting in high-quality training corpora for NER tasks.
- Built on mBERT architecture with 177M parameters
- Implements transformer-based token classification
- Supports TensorBoard integration
- Uses Safetensors for model storage
Core Capabilities
- Multilingual NER across 9 European languages
- Robust performance on encyclopedic content
- Easy integration with Hugging Face transformers pipeline
- Grouped entity recognition support
Frequently Asked Questions
Q: What makes this model unique?
This model combines neural and knowledge-based approaches for silver data creation, achieving up to 6 span-based F1-score points improvement over previous state-of-the-art systems. It's particularly effective for encyclopedic content while supporting multiple languages simultaneously.
Q: What are the recommended use cases?
The model is best suited for multilingual NER tasks, particularly with Wikipedia-style content. For optimal results in other domains, it's recommended to combine WikiNEuRal with domain-specific datasets like CoNLL.