Aspect-Based Opinion Mining Framework for Product Rating Embedded with Fuzzy Decision

Abstract

The technique of evaluating text on a subject jotted down in natural language and classifying it with polarity as neutral form, positive form, or negative form based on the sentiments, emotions, and expressions by humans in it is known as opinion mining or sentiment analysis. Analyzing and extracting opinions from such a large volume of reviews manually is probably impossible. A self-dependent automated opinion mining strategy is required to resolve this issue. It can primarily be carried out at three levels: document, sentence, and aspect level. Work proposed is an effort to rate a product and predict its demand based on features extracted from aspects of opinion expressed with fuzzy set decision boundary. Aspect identification, aspect-based opinion word identification, and inclination as positive, negative or neutral are the three major tasks in aspect-based opinion mining. Weights were assigned according to the polarity percentage of the words extracted. Polarity scores of the labeled dataset were then divided into 5 fuzzy set decision boundaries, confirmed positive, positive, neutral, negative, and confirmed negative, to accurately predict the correlation between polarity score obtained and product rating. Algorithm proposed was operated on aspect-based corpus for product rating which acquired from GitHub, comprised of 7563 rows of product aspects; the analysis suggests that maximum share 44% lies with positive range of fuzzy boundary, and this can be tapped for trend prediction of customer choice; the algorithm successfully provides a fuzzy-assisted prediction mechanism for product rating for the corpus with labeled aspects; algorithm performs at par with existing technique, and it also provides percentage share of products presence in market for future trends.