A Deep Learning-Fuzzy Based Hybrid Ensemble Approach for Aspect Level Sentiment Classification

Abstract

Aspect level sentiment classification (ALSC) has gained high importance in the era of an e-commerce-based economy. It allows manufacturers to improve the designs of their products based on users’ feedback. However, only a few datasets of limited domains are available for ALSC task. To push forward the research in automated ALSC, this study contributes car dataset of the automobile domain. In this study, a novel fuzzy ensemble technique is also proposed based on the mathematical analysis of confidence scores of base deep neural networks. The proposed approach allows for the correction of the misclassifications of base deep learners through a reward and penalization strategy. The experimental results on five benchmark datasets show that the proposed approach outperforms the constituent base deep neural networks and several other important baselines. The proposed Fuzzy ensemble also performed at par with the most recent Graph Convolution Neural Networks on the basis of Friedman and Nemenyi Tests.