Bayesian game model based unsupervised sentiment analysis of product reviews

Abstract

Sentiment Analysis is a task of computationally recognizing and contextualizing opinions stated in a text. We mainly assess whether the writer's attitude towards a specific topic, or a product, is positive, negative, or neutral. Numerous machine learning and fuzzy logic methods have been reconnoitered for sentiment analysis. Yet, the application of mathematical optimization techniques for sentiment tagging is still unexplored. This study presents a novel mathematical framework for sentiment analysis of reviews based on Game Theory. We identify whether the sentiment of a review is positive or negative. In the first step, we comprehend a review and derive context scores from review comments using the SentiWordNet lexicon. We comprehensively combine the computed context and rating scores using the Bayesian Game Model to deduce the sentiment of reviews. Experimental results on three benchmark review datasets, viz. Food, Mobile, and Electronics demonstrate that the proposed model yields state-of-the-art results. We also statistically validated the stability and correctness of the results. The proposed model ensures rational and consistent results. The utility of the game theory model for sentiment analysis creates a new paradigm for diverse NLP tasks.