Sarcasm Detection in News Headlines Using ML and DL Models

Sarcasm is a form of language defined using words or phrases that express the opposite of their actual meaning, often with a purpose of mocking or criticizing something or someone. Sarcasm detection is crucial for false news detection, opinion mining, sentiment analysis, detecting cyberbullies, online trolls, and other similar activities. Detecting Sarcasm is a part of Sentimental Analysis. This paper focuses on analysis of news headline to detect sarcasm using ensemble Machine Learning models like XGBoost, AdaBoostand Deep learning models like BiLSTM, CNN, RNNand a Hybrid CNN and BiLSTM model. The RNN model outperformed all of the other models with an accuracy of 0.79 and balanced F1 score of 0.76, which indicates its proficiency in discerning sarcastic content. LIME analysis is implemented to evaluate contribution of each word in a news headline towards sarcasm.