Probabilistic Model Based Context Augmented Deep Learning Approach for Sarcasm Detection in Social Media

Abstract

Sarcasm is an implicit form of sentiment widely used on social media which conveys the opposite meaning of what is actually conveyed. It is usually aimed at mocking or criticizing someone or some-thing. Identification of sarcasm is harder even for human-beings due to its intrinsic ambiguous nature. The commonly used Sentiment Analysis system automatically identifies the polarity of the contents, but they fail to consider the effect of sarcastic statements. System’s performance reduces when they misinterpret the sarcastic statements. Therefore, automatic recognition of sarcastic statements on data collected from social networks can make the existing sentiment analysis system and other applications based on NLP perform better. In this paper, a probabilistic model-based context augmentation is proposed for detection of sarcasm from Twitter. The proposed work operates in two phases: In the first phase, a probabilistic model has been proposed which computes the confidence level of the tweets. Confidence level denotes the probability that a tweet can be considered sarcastic.  Based on the confidence level the overall weightage score for the terms pairs would be generated. In the second phase, the proposed probabilistic weightage score would be augmented with the context information obtained from the word-to-vector model. The augmented context set is then passed on to Convolutional neural network in order to indentify sarcasm. The main aim of the proposed work is to provide an augmented context in addition to the existing contextual information available in the conventional methods. The proposed approach attained an accuracy of 97.25% which is a significant improvement over the 85.26% achieved by the traditional contextual approach.