An Approach to Detect Sentence Level Sarcasm Using Deep Learning Techniques

Abstract

Today is the era of customer self-service, where people use conversation agents (chatbots) to get their query solved in minimum time and cost. Use of conversation Agent gives real time experience to user/customer to get answers very fast. To make this experience more genuine it needs understanding of human emotions and which is the most complex task to perform as facial expression and verbal details are unavailable. Most of data over the internet is in textual format which needs to process to get required answers that is why one of the most popular area in natural language processing is Sentiment analysis which focuses on solving this issue. Sarcasm is critical sentiment which is very difficult to recognize by machines. User express sarcasm to show their anger, disagreement using positive words over internet forums, social media and over shopping sites for reviews about product, services, situation, workplace etc. This paper gives a combined approach by extracting pragmatic features like emoticons, use of hyperbole, punctuations and special words used in sentence to detect sentence level sarcasm using deep learning techniques such as LSTM which will help machine for better understanding of natural language.