BCMS-BERTić NER Model
| Property | Value |
|---|---|
| Parameter Count | 110M |
| License | Apache 2.0 |
| Languages | Bosnian, Croatian, Montenegrin, Serbian |
| Task Type | Named Entity Recognition |
| F1 Score | 91.38 |
What is bcms-bertic-ner?
BCMS-BERTić-NER is a specialized transformer-based model designed for Named Entity Recognition in Balkan languages. It's a fine-tuned version of the BERTić base model, specifically optimized to identify and classify named entities (PER, LOC, ORG, MISC) in Bosnian, Croatian, Montenegrin, and Serbian texts. The model builds upon extensive training data from multiple datasets, including hr500k, SETimes.SR, and ReLDI collections.
Implementation Details
The model utilizes the ELECTRA architecture and has been trained on a diverse collection of datasets totaling over 768,000 tokens. The training data includes both standard language sources and internet-based content (Twitter), ensuring robust performance across different text styles.
- Built on the ELECTRA architecture with 110M parameters
- Trained on four major datasets covering different language variants
- Supports missing diacritics handling
- Implements BIO tagging scheme for entity recognition
Core Capabilities
- Named Entity Recognition for four entity types: Person (PER), Location (LOC), Organization (ORG), and Miscellaneous (MISC)
- High-performance entity detection with 91.38 F1 score
- Handling of both formal and informal language variants
- Support for multiple Balkan language variants
Frequently Asked Questions
Q: What makes this model unique?
The model's uniqueness lies in its specialized focus on Balkan languages and its ability to handle multiple language variants with high accuracy. It's particularly notable for its handling of missing diacritics and support for both standard and internet language varieties.
Q: What are the recommended use cases?
The model is ideal for applications requiring named entity recognition in Balkan languages, such as information extraction, content analysis, and automated text processing systems. It's particularly suited for both formal documents and social media content analysis.
