A Novel Approach to Feature Hierarchy in Aspect Based Sentiment Analysis Using OWA Operator

Abstract

In today’s digital age opinions enable effective decision-making ability of both the customer and manufacturer. These opinions can be analysed using aspect-based sentiment analysis. The analysis uses the features of the product for determining the final polarity score. Researchers have focussed on extracting the features and giving them optimum weights (selected top k features) for final classification. A major drawback with these approaches is that they follow a uniform weight assignment scheme for all the features. It is observed that in any domain there can be a ‘three-class-bifurcation’ (either implicit or explicit)—low level, medium level, high level of features. In real-life the features belonging to high level are sometimes adapted at either low or medium level. This can be termed as overlapping of features. This encourages us to design a new feature hierarchy for effective weight distribution of overlapped features. The proposed methodology comprises of two steps: lookup and update. The resultant updated weights of the features can then be used for further processing through classification in aspect-based sentiment analysis.