HATE-ITA
| Property | Value |
|---|---|
| Developer | MilaNLProc |
| License | GNU GPLv3 |
| Performance | F1 Score: 0.83 |
| Base Architecture | XLM-T |
What is hate-ita?
HATE-ITA is a specialized binary hate speech classification model designed specifically for Italian social media text. Developed by researchers Debora Nozza, Federico Bianchi, and Giuseppe Attanasio, it addresses the critical need for hate speech detection in non-English languages. The model leverages multi-language training data, combining large English datasets with available Italian resources to achieve superior performance compared to monolingual alternatives.
Implementation Details
The model is built upon the XLM-T architecture and is available in multiple variants including base and large versions. It can be easily implemented using the Transformers library from Hugging Face, making it accessible for both researchers and practitioners.
- Built on XLM-T architecture
- Multiple model variants available (base and large)
- Simple integration with Transformers pipeline
- Achieves 0.83 F1 score on test set
Core Capabilities
- Binary classification of hate speech in Italian text
- Effective processing of social media content
- Cross-lingual transfer learning benefits
- Adaptation to language-specific slurs
Frequently Asked Questions
Q: What makes this model unique?
HATE-ITA stands out for its multi-language training approach, combining English and Italian datasets to achieve better performance than monolingual models. It specifically addresses the gap in hate speech detection for the Italian language while maintaining high accuracy.
Q: What are the recommended use cases?
The model is primarily designed for detecting hate speech in Italian social media text. However, practitioners should note the ethical considerations mentioned by the authors regarding potential limitations in maintaining consistent precision across different targets and categories.
