Sarcasm Detection in Newspaper Headlines

Abstract

Sarcasm is an important part of communication, and detecting sarcasm is difficult for humans, let alone computers. Newspapers often seem to employ sarcasm in their headlines to grab the readers' attention. However, more often than not, the readers find it difficult to detect the irony in the headlines, thus getting a wrong idea about that particular news and further passing on their understanding to their friends, colleagues, etc. Thus, a system which can automatically and reliably detect sarcasm is more important now than ever. We build sarcasm detectors using neural networks and attempt to understand how a computer learns the patterns of sarcasm. The input to our project consists of sequences that are labeled sarcastic or non-sarcastic. These sequences come from two different datasets containing news headlines and social media commentary. Our classifiers are evaluated on their accuracies. Our model performs highly and is capable of reliably classifying sarcastic or non-sarcastic phrases.