Aspect-Level Sentiment Analysis through Aspect-Oriented Features

Abstract

Aspect-level sentiment analysis is essential for businesses to comprehend sentiment polarities associated with various aspects within unstructured texts. Although several solutions have been proposed in recent studies in sentiment analysis, a few challenges persist. A significant challenge is the presence of multiple aspects within a single written text, each conveying its own sentiments. Besides this, the exploration of ensemble learning in the existing literature is limited. Therefore, this study proposes a novel aspect-level sentiment analysis solution that utilizes an ensemble of Bidirectional Long Short-Term Memory (BiLSTM) models. This innovative solution extracts aspects and sentiments and incorporates a rule-based algorithm to combine accurate sets of aspect and sentiment features. Experimental analysis demonstrates the effectiveness of the proposed methodology in accurately extracting aspect-level sentiment features from input texts. The proposed solution was able to obtain an F1 score of 92.98% on the SemEval-2014 Restaurant dataset when provided with the correct set of aspect-level sentiment features and an F1 score of 95.54% on the SemEval-2016 Laptop dataset when provided with the aspect-level sentiment features generated by the aspect-sentiment mapper algorithm. Doi: 10.28991/HIJ-2024-05-01-09 Full Text: PDF

Keywords

• aspect-level sentiment analysis
• ensemble model
• deep learning.