UPB at SemEval-2021 Task 5: Virtual Adversarial Training for Toxic Spans Detection

TL;DR

This paper presents a semi-supervised approach using Virtual Adversarial Training with Transformer models and CRFs for toxic span detection, achieving state-of-the-art F1-scores in SemEval-2021 Task 5.

Contribution

It introduces the application of Virtual Adversarial Training in a semi-supervised setting for toxic span detection with Transformer models, improving robustness and performance.

Findings

Achieved an F1-score of 65.73% in official submission.

Improved F1-score to 66.13% after post-evaluation tuning.

Demonstrated the effectiveness of Virtual Adversarial Training for toxicity detection.

Abstract

The real-world impact of polarization and toxicity in the online sphere marked the end of 2020 and the beginning of this year in a negative way. Semeval-2021, Task 5 - Toxic Spans Detection is based on a novel annotation of a subset of the Jigsaw Unintended Bias dataset and is the first language toxicity detection task dedicated to identifying the toxicity-level spans. For this task, participants had to automatically detect character spans in short comments that render the message as toxic. Our model considers applying Virtual Adversarial Training in a semi-supervised setting during the fine-tuning process of several Transformer-based models (i.e., BERT and RoBERTa), in combination with Conditional Random Fields. Our approach leads to performance improvements and more robust models, enabling us to achieve an F1-score of 65.73% in the official submission and an F1-score of 66.13% after further tuning during post-evaluation.