Sefamerve ARGE at SemEval-2021 Task 5: Toxic Spans Detection Using Segmentation Based 1-D Convolutional Neural Network Model

AbstractThis paper describes our contribution to SemEval-2021 Task 5: Toxic Spans Detection. Our approach considers toxic spans detection as a segmentation problem. The system, Waw-unet, consists of a 1-D convolutional neural network adopted from U-Net architecture commonly applied for semantic segmentation. We customize existing architecture by adding a special network block considering for text segmentation, as an essential component of the model. We compared the model with two transformers-based systems RoBERTa and XLM-RoBERTa to see its performance against pre-trained language models. We obtained 0.6251 f1 score with Waw-unet while 0.6390 and 0.6601 with the compared models respectively.