TOBB ETU at CheckThat! 2021: Data engineering for detecting check-worthy claims

Abstract

In this paper, we present our participation in CLEF 2021 CheckThat! Lab's Task 1 on check-worthiness estimation in tweets. We explore how to fine-tune transformer models effectively by changing the train set. The methods we explore include language-specific training, weak supervision, data augmentation by machine translation, undersampling, and cross-lingual training. As our primary model submitted for official results, we fine-tune language-specific BERT-based models using cleaned tweets for each language. Our models ranked 1st in Spanish and Turkish datasets. However, our rank in Arabic, Bulgarian, and English datasets is 6t?, 4t?, and 10th, respectively. © 2021 Copyright for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).