OPTIMIZATION TECHNIQUES BASED SENTIMENT ANALYSIS

Abstract

Sentiment Analysis is a task under the domain of Natural Language Processing that plays a crucial role in understanding and quantifying emotions, opinions, and attitudes. The abundance of online data drives businesses to leverage sentiment analysis as a means to monitor and gauge consumer sentiments and emotions, enabling them to make informed decisions and tailor their services to meet customer needs. Many existing approaches heavily rely on machine learning, necessitating large datasets for pre-training and incurring significant computational complexity. To tackle this issue, we propose unsupervised sentiment classification models for sentiment analysis. This thesis introduces frameworks based on mathematical optimization techniques namely game theory and Multi Criteria Decision Making. The integration of these mathematical techniques generates robust algorithms for sentiment tagging. We use textual feedback and star ratings of reviews and apply mathematical optimization techniques to deduce the correct sentiment for the reviews. In the thesis, we have performed binary and tertiary classification of review comments on datasets of varied domains. We have also introduced two explicit models for sentiment analysis of Hindi review comments. This assures that the mathematical optimization techniques with minor modifications can adapt to any language. We have also addressed the negation handling challenge of the sentiment analysis. To ascertain the relevance of the sentiment analysis task, we used it to generate two recommendation models, which produce promising results. In summary, our novel unsupervised sentiment classification models present effective solutions to the challenges posed by the vast amounts of online data and the resource intensive nature of conventional machine learning approaches. By utilizing mathematical optimization models, we offer efficient, scalable, and accurate sentiments of written reviews. Furthermore, the models guarantee logical and consistent outcomes, instilling confidence in the accuracy of sentiment classifications.