Sarcasm Detection in Newspaper Headlines

In the current age of information saturation, where headlines serve as primary points of engagement, the pervasive use of sarcasm in news headlines poses a challenge for readers’ comprehension, failing to discern sarcasm may lead to misunderstandings, misinformation dissemination, and harm journalistic integrity. This study addresses this critical issue and enhances the credibility of news sources by employing various ensemble learning algorithms and deep learning models with different types of word embedding, the evaluation includes various possible performance metrics and a complete correlation of the outcomes. XGBoost consistently outperforms other models with higher accuracy, precision, recall, and $F 1$-score in sarcasm detection. Among deep learning models, BiLSTM emerges as the top performer, achieving 0.86 accuracy, capturing nuanced language cues effectively and appears as the optimal model for sarcasm detection. BERT stands out with superior F1 scores, showcasing its exceptional ability in deciphering sarcastic language nuances.