Sarcasm Detection in Tweets as Contrast Sentiment in Words Using Machine Learning and Deep Learning Approaches

Abstract

Sarcasm is a remark that clearly means the opposite of what is said, made to criticize something in a humorous way. The main goal of the proposed work is to identify sarcasm in plain text. Sarcasm detection in a text is a difficult task due to lack of context of the tweet, lack of user’s character and personality and lack of expression and body language. We present a novel method in feature engineering based on the contrast in phrases of the sarcastic sentence. It can recognize the positive phrases followed by negative situation phrase (example- I can’t wait for the algebra exam tomorrow!) and the inverse of it in sarcastic tweets. It has been done by getting the sentiment of various parts of the tweet by dividing the tweet. Features to know the context of words used in a sentence were extracted by many different methods. We used a neural network model using the ReLU activation function for the detection of sarcasm in tweets, which improves the f1-score as compared to machine learning approaches. It is found that the neural network model outperforms the machine learning model in terms of accuracy and f1-score.