ner-dutch-large
| Property | Value |
|---|---|
| Author | flair |
| F1-Score | 95.25% (CoNLL-03 Dutch) |
| Framework | Flair |
| Base Model | XLM-RoBERTa-large |
| Paper | FLERT: Document-Level Features for Named Entity Recognition |
What is ner-dutch-large?
ner-dutch-large is a state-of-the-art Named Entity Recognition model specifically designed for Dutch language processing. Built using the FLERT architecture, it leverages document-level XLM-R embeddings to achieve exceptional accuracy in identifying four distinct types of entities: person names (PER), locations (LOC), organizations (ORG), and miscellaneous names (MISC).
Implementation Details
The model is implemented using Flair framework and utilizes TransformerWordEmbeddings with XLM-RoBERTa-large as its backbone. It employs a sophisticated architecture without CRF or RNN layers, focusing instead on document-context aware predictions through fine-tuned transformer embeddings.
- Uses document-level context for improved entity recognition
- Implements first-subtoken pooling strategy
- Trained with AdamW optimizer and OneCycleLR scheduler
- Employs 256-dimensional hidden states
Core Capabilities
- High-accuracy entity recognition with 95.25% F1-score
- Four-class classification system (PER, LOC, ORG, MISC)
- Document-aware context processing
- Native integration with Flair framework
- Efficient processing of Dutch text
Frequently Asked Questions
Q: What makes this model unique?
The model's uniqueness lies in its use of document-level features through the FLERT architecture, combined with state-of-the-art performance on Dutch NER tasks. The integration of XLM-RoBERTa-large embeddings and document context awareness sets it apart from traditional NER models.
Q: What are the recommended use cases?
This model is ideal for Dutch text processing applications requiring named entity recognition, such as information extraction, document analysis, and automated content categorization. It's particularly effective for applications requiring high-accuracy identification of persons, locations, organizations, and miscellaneous named entities in Dutch text.
