UNSUPERVISED TEXTUAL SARCASM DETECTION USING OPTIMIZATION TECHNIQUES

Abstract

Sarcasm detection is a form of figure of speech that conveys the opposite of its literal meaning, often to express insult, wit, irritation, or ridicule. In text, sarcasm is typically conveyed through positive or intensified positive words to mask negative feelings. With the rise of social media platforms like Twitter, Facebook, and WhatsApp, posting sarcastic messages has become a common way to avoid direct negativity. However, detecting these indirect negativities is crucial as they significantly impact businesses. The challenge in analysing sarcasm lies in the gap between its literal and intended meanings. Despite extensive research in natural language processing (NLP) and sarcasm detection, there is a notable lack of comparative analysis among different NLP techniques and their ability to correctly classify sarcastic content. Additionally, there is a scarcity of studies on using mathematical optimization techniques for sarcasm detection and a neglect of the intonation and tonal traits of sarcasm. This thesis addresses these gaps by introducing frameworks that integrate mathematical optimization techniques with NLP models. These frameworks generate robust algorithms for detecting sarcasm and its inherent tonal nature. We utilize sentence scores from sentiment lexicon models and apply mathematical optimization techniques to identify sarcasm in social media comments. The thesis includes binary and tertiary classification of social media comments across various domains. It also presents a model for detecting sarcasm in Hindi comments, demonstrating that these mathematical optimization techniques can be adapted to any language with minor modifications. We have specifically focused on incorporating the tonal traits of sarcasm into sentiment analysis. The primary objective is to expand knowledge in this area and provide new perspectives on the strengths and weaknesses of the proposed models. This research aims to contribute to both the academic community and companies that develop or use this technology. Our study employs a qualitative approach supported by quantitative data. An extensive literature review was conducted to deepen our understanding of the field. Benchmark datasets were used for analysis, and the results form the basis for evaluating the selected models ability to identify sarcasm based on metrics such as accuracy, precision, recall, and F1 score. The results indicate that the proposed mathematical optimization-based models are effective for classifying and detecting sarcasm. These models offer efficient, scalable, and accurate solutions for analysing written reviews by leveraging mathematical optimization techniques. iv In summary, our novel unsupervised sarcasm detection methods provide effective solutions to the challenges posed by large amounts of online data and the resource-intensive nature of conventional machine learning approaches. By utilizing mathematical optimization models, we ensure logical and consistent outcomes, thereby enhancing confidence in the accuracy of sarcasm classifications. These models are designed to be efficient, scalable, and accurate in detecting sarcasm in written contexts.