roberta-large-ner-english
| Property | Value |
|---|---|
| Parameter Count | 354M |
| License | MIT |
| Framework | PyTorch, TensorFlow, ONNX |
| Training Data | CoNLL-2003 |
What is roberta-large-ner-english?
roberta-large-ner-english is a specialized Named Entity Recognition (NER) model fine-tuned from RoBERTa-large on the CoNLL-2003 dataset. It's particularly notable for its exceptional performance on informal text like emails and chat data, especially excelling at recognizing entities that don't begin with uppercase letters.
Implementation Details
The model was trained on the CoNLL-2003 dataset, utilizing 17,494 training samples and 3,250 validation samples. It identifies four main entity types: Person (PER), Organization (ORG), Location (LOC), and Miscellaneous (MISC). Unlike traditional implementations, this model simplifies the labeling scheme by removing B- and I- prefixes.
- Achieves 97.53% overall F1 score on CoNLL-2003 validation set
- Outperforms Spacy's transformer model on informal text
- Supports multiple deep learning frameworks including PyTorch and TensorFlow
- Implements efficient token classification with RoBERTa architecture
Core Capabilities
- Person name recognition (99.20% F1 score)
- Organization detection (96.44% F1 score)
- Location identification (98.28% F1 score)
- Miscellaneous entity recognition (92.77% F1 score)
- Superior performance on informal communication channels
Frequently Asked Questions
Q: What makes this model unique?
This model stands out for its exceptional performance on informal text and entities without capitalization, making it particularly suitable for processing emails and chat communications. It consistently outperforms other models in these scenarios, showing significant improvements over established solutions like Spacy.
Q: What are the recommended use cases?
The model is ideal for: Email analysis and signature detection, Chat message processing, Social media content analysis, and any NER task involving informal text where traditional capitalization rules might not apply. It's particularly effective for business applications requiring robust entity extraction from communication data.
