Named Entity Recognition System

Create a Named Entity Recognition system for custom entity types.

Task details:
- Entity types: [LIST YOUR ENTITIES]
- Domain: [MEDICAL/LEGAL/FINANCE/GENERAL]
- Language: [LANGUAGE]
- Text type: [DOCUMENTS/SOCIAL MEDIA/etc.]

NER requirements:
1. Data preparation:
   - Annotation format (BIO/BIOES)
   - Tokenization alignment
   - Label encoding
2. Model architecture:
   - BERT-based token classification
   - BiLSTM-CRF option
   - Span-based models
3. Training:
   - Handling long documents
   - Class imbalance strategies
   - Cross-validation
4. Evaluation:
   - Entity-level F1
   - Strict vs. lenient matching
   - Error analysis by entity type
5. Post-processing:
   - Entity linking
   - Coreference resolution
   - Confidence thresholding
6. Inference:
   - Batch processing
   - Real-time extraction
   - Output formatting
7. Active learning:
   - Uncertainty sampling
   - Annotation interface

Use Hugging Face Transformers or spaCy.