fact-or-opinion-xlmr-el
| Property | Value |
|---|---|
| License | Apache 2.0 |
| Languages | English, Greek, Multilingual |
| Architecture | XLM-RoBERTa-base |
| Test Accuracy | 95.2% |
What is fact-or-opinion-xlmr-el?
fact-or-opinion-xlmr-el is a sophisticated binary classification model developed through collaboration between the Hellenic Army Academy and Technical University of Crete. Built on XLM-RoBERTa architecture, it distinguishes between factual and opinion-based statements in both English and Greek, with zero-shot capabilities for other languages.
Implementation Details
The model was trained on a dataset of approximately 9,000 annotated sentences, originally in English and translated to Greek using Google Translate. Training occurred over 5 epochs with a batch size of 8, achieving impressive metrics including 95.2% accuracy, 94.5% precision, and 96% recall.
- Binary classification head on XLM-RoBERTa-base
- Label 0 indicates Opinion/Subjective sentences
- Label 1 represents Fact/Objective sentences
- Supports cross-lingual inference through zero-shot learning
Core Capabilities
- Accurate classification of statements as facts or opinions
- Native support for English and Greek languages
- Zero-shot learning capabilities for other languages
- High-performance metrics across precision, recall, and F1-score
- Suitable for both academic and practical applications
Frequently Asked Questions
Q: What makes this model unique?
The model's distinctive feature is its bilingual training approach combining English and Greek, while maintaining high accuracy (95.2%) and offering zero-shot capabilities for other languages supported by XLM-R.
Q: What are the recommended use cases?
The model is ideal for content analysis, media monitoring, educational applications, and research purposes where distinguishing between factual and opinion-based content is crucial. It's particularly valuable for organizations working with English and Greek content, though it can be applied to other languages through zero-shot learning.
