Akademik Veri Yönetim Sistemi
33rd IEEE Conference on Signal Processing and Communications Applications, SIU 2025, İstanbul, Türkiye, 25 - 28 Haziran 2025, (Tam Metin Bildiri)
Irony detection in Turkish texts is a challenging task due to the nuanced nature of irony in natural language. The limited availability of irony data for Turkish is also an important limitation to train models. In this study, the effect of data augmentation with translation for irony detection is investigated, and new models are proposed using ensemble and meta-learning approaches, in order to increase the detection performance. In the first stage, performance of classical machine learning and transformer-based models is investigated with the augmented dataset. Subsequently, ensemble and meta-learning models are constructed with variations of the transformer-based model trained on different data subsets. The created models are evaluated under cross-validation, and the performance of each classifier is measured by accuracy, precision, recall, and F1 score. The results show that the ironic sentences obtained with translation do not have exactly the same structure and quality as the original ones. However, proposed ensemble and meta learning-based models trained with the augmented data improve the irony detection performance and carry potential to be used.