NuNER_Zero
| Property | Value |
|---|---|
| License | MIT |
| Framework | PyTorch |
| Paper | View Paper |
| Downloads | 193,442 |
What is NuNER_Zero?
NuNER_Zero is a state-of-the-art zero-shot Named Entity Recognition (NER) model that leverages the GLiNER architecture. It represents a significant advancement in entity recognition, achieving a 3.1% improvement in token-level F1-Score compared to GLiNER-large-v2.1 on standard benchmarks. The model was trained on the NuNER v2.0 dataset, which combines carefully curated subsets of Pile and C4 annotated using LLMs.
Implementation Details
The model operates as a token classifier, enabling the detection of arbitrarily long entities. It requires input in the form of concatenated entity types and text, with the unique requirement that all labels must be lowercase.
- Built on GLiNER architecture for flexible entity recognition
- Trained on LLM-annotated NuNER v2.0 dataset
- Implements token classification for handling variable-length entities
- Includes entity merging capabilities for continuous text spans
Core Capabilities
- Zero-shot named entity recognition across various entity types
- Support for custom entity type definitions
- Efficient processing of long text sequences
- High accuracy in entity boundary detection
- Simple integration through the gliner Python package
Frequently Asked Questions
Q: What makes this model unique?
NuNER_Zero stands out for its zero-shot capabilities and superior performance, offering a 3.1% improvement over existing models. It uniquely combines token classification with the GLiNER architecture, allowing for more flexible entity recognition without requiring training data for new entity types.
Q: What are the recommended use cases?
The model is ideal for applications requiring dynamic entity recognition where training data isn't available for specific entity types. It's particularly useful in scenarios involving custom entity types, document processing, and automated text analysis systems.
