Small Electra for Hate: A Transformer Lightweight Approach to Hate Text Classification

Lazo Vasquez R.; Ochoa-Luna J. · Communications in Computer and Information Science (2025) · DOI: 10.1007/978-3-031-91428-7_18

Resumen

Detecting hate speech in text is essential in today’s digital age, where online platforms strive to eliminate toxic discourse and create a healthy environment for users. In response to this challenge, we introduce SE4Hate, a Transformer-based model built on Electra Small, designed to tackle hate speech classification on social networks. Our model achieves an F1 score of 78.8785%, surpassing conventional approaches based on Transformers and Large Language Models (LLMs) like RoBERTa, ChatGPT, and FastChat T5, thanks to its cost-effective architecture and strong performance. Despite trailing the top state-of-the-art approach by 0.6%, which relies on an architecture seven times more complex, the model still provides notable improvements in cost-efficiency while maintaining a lightweight, traditional Transformer-based design. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.

Instituciones: Universidad Católica San Pablo