Sarcasm Detection In News Headlines Using ML And DL Models

Abstract

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.