UPB at SemEval-2020 Task 12: Multilingual Offensive Language Detection on Social Media by Fine-tuning a Variety of BERT-based Models

TL;DR

This paper explores multilingual offensive language detection on social media using fine-tuned BERT-based models across five languages, achieving competitive results in the SemEval-2020 shared task.

Contribution

It evaluates multiple transformer architectures and training strategies for multilingual offensive language detection, providing insights into their comparative effectiveness.

Findings

Best models achieved top 10 rankings in several languages.

Multilingual models performed competitively with single-language models.

Fine-tuning on combined datasets improved detection accuracy.

Abstract

Offensive language detection is one of the most challenging problem in the natural language processing field, being imposed by the rising presence of this phenomenon in online social media. This paper describes our Transformer-based solutions for identifying offensive language on Twitter in five languages (i.e., English, Arabic, Danish, Greek, and Turkish), which was employed in Subtask A of the Offenseval 2020 shared task. Several neural architectures (i.e., BERT, mBERT, Roberta, XLM-Roberta, and ALBERT), pre-trained using both single-language and multilingual corpora, were fine-tuned and compared using multiple combinations of datasets. Finally, the highest-scoring models were used for our submissions in the competition, which ranked our team 21st of 85, 28th of 53, 19th of 39, 16th of 37, and 10th of 46 for English, Arabic, Danish, Greek, and Turkish, respectively.