bert-base-NER
| Property | Value |
|---|---|
| Parameter Count | 108M |
| License | MIT |
| Paper | Link |
| F1 Score | 91.3% (test) |
What is bert-base-NER?
bert-base-NER is a fine-tuned BERT model specifically designed for Named Entity Recognition (NER). Built on the bert-base-cased architecture, this model has been trained to identify four types of entities: Location (LOC), Organizations (ORG), Person (PER), and Miscellaneous (MISC). The model achieves state-of-the-art performance on the CoNLL-2003 Named Entity Recognition task.
Implementation Details
The model was trained on the English version of the CoNLL-2003 dataset, which contains Reuters news stories. Training was performed on a single NVIDIA V100 GPU using hyperparameters recommended in the original BERT paper. The model utilizes a token classification approach to identify entity boundaries and types.
- Trained on 203,621 tokens across 14,987 sentences
- Achieves 92.11% precision and 93.06% recall
- Supports both beginning (B-) and inside (I-) tags for entity identification
Core Capabilities
- Real-time named entity recognition
- Support for nested entity detection
- Easy integration with Transformers pipeline
- Handles both single sentences and document-level text
Frequently Asked Questions
Q: What makes this model unique?
This model combines the powerful BERT architecture with specific NER training, achieving a 91.3% F1 score on the test set. It's particularly notable for its balanced performance across different entity types and its ability to handle cased text.
Q: What are the recommended use cases?
The model is ideal for information extraction tasks, news analysis, document processing, and any application requiring the identification of persons, organizations, locations, and miscellaneous entities in English text. It's particularly well-suited for news and formal content processing.
