Sarcasm Detection in News Headlines with Deep Learning

Abstract

Sarcasm detection is one of the recent topics studied in the field of natural language processing. Although sarcasm detection is generally carried out through social media comments in the literature, it can also be applied to news headlines that are expected to be completely objective and reflect reality. In this study, sarcasm detection was carried out using various deep learning models in a dataset containing sarcastic and non-sarcastic news headlines. The accuracy of classification results of BERT, RNN, LSTM, and GRU models and their training time performance were compared. While the BERT model reached the highest accuracy (0.88), RNN was the most successful model in terms of training time performance.