BCMS-BERTIC-ParlaSent-BCS-TER
| Property | Value |
|---|---|
| Author | CLASSLA |
| Task | Ternary Sentiment Classification |
| Performance | 0.7941 ± 0.0101 Macro F1 |
| Paper | ParlaSent-BCS Dataset Paper |
What is bcms-bertic-parlasent-bcs-ter?
This is a specialized transformer-based model designed for sentiment analysis in Bosnian, Croatian, Montenegrin, and Serbian languages. It's based on the BCMS-BERTIC architecture and fine-tuned on the BCS Political Sentiment dataset to perform ternary classification (Negative, Neutral, Positive) on parliamentary debates.
Implementation Details
The model was fine-tuned using simpletransformers library with 9 training epochs, determined through hyperparameter optimization. It significantly outperforms other models like EMBEDDIA/crosloengual-bert and XLM-RoBERTa-base in sentiment classification tasks.
- Fine-tuned on sentence-level political sentiment data
- Achieves state-of-the-art performance with 0.7941 macro F1 score
- Implements three-way classification: Negative, Neutral, Positive
Core Capabilities
- Accurate sentiment analysis for South Slavic languages
- Specialized for political discourse analysis
- Easy integration with simpletransformers library
- Robust performance across different parliamentary contexts
Frequently Asked Questions
Q: What makes this model unique?
The model specializes in South Slavic languages and achieves state-of-the-art performance in political sentiment analysis, significantly outperforming multilingual alternatives like XLM-RoBERTa.
Q: What are the recommended use cases?
This model is ideal for analyzing political discourse, parliamentary debates, and general sentiment analysis in Bosnian, Croatian, Montenegrin, and Serbian languages. It's particularly useful for researchers and analysts working with political texts in these languages.
