spaCy NER Skill
Capabilities
- Train custom spaCy NER models
- Configure entity extraction pipelines
- Design annotation schemas
- Implement entity linking
- Set up model evaluation
- Deploy efficient NER inference
Target Processes
- entity-extraction-slot-filling
- chatbot-design-implementation
Implementation Details
spaCy Components
- NER: Named Entity Recognition
- EntityLinker: Link to knowledge bases
- EntityRuler: Rule-based matching
- SpanCategorizer: Overlapping entities
Training Configuration
- config.cfg setup
- Training data format (spaCy v3)
- Augmentation strategies
- Evaluation metrics
Configuration Options
- Base model selection (en_core_web_*)
- Custom entity types
- Training parameters
- GPU acceleration
- Model packaging
Best Practices
- Quality annotation data
- Balance entity types
- Use prodigy for annotation
- Regular model evaluation
Dependencies
- spacy
- spacy-transformers (optional)