Sentiment Analysis Using Emotion‐Guided Polarized Capsule Dropout With Bidirectional Long Short‐Term Memory on User Review Data

ABSTRACT

Sentiment Analysis (SA) is an essential task in Natural Language Processing (NLP) to discriminate emotions and opinions expressed in text. Though existing algorithms on sentiment analysis struggle to capture hierarchical semantic structure and contextual dependencies in text data. To address these limitations, this research paper developed Emotion‐Guided Polarized Capsule Dropout with Bidirectional Long Short‐Term Memory (EG‐PCD‐BiLSTM) for effective sentiment analysis. Emotion‐guided routing with a Capsule network is incorporated in BiLSTM to preserve semantic relationships and improve the feature representation of text data. The emotion‐guided routing improves capsule networks by including emotion scores from lexicons. It directed the emotionally rich tokens towards suitable sentiment capsules, enhancing subtle classification. The Polarized Capsule Dropout (PCD) selectively deactivates the low‐confidence capsules based on vector magnitude. This process preserves meaningful feature representation and enhances model generalization by filtering out the noisy features. Moreover, the Bi‐LSTM captures both past and future contextual features in sequential data. The EG‐PCD‐BiLSTM model obtains the highest accuracy of 97.94% on the Amazon review dataset and 99.03% on the IMDB dataset when compared to existing algorithms. The experimental outcomes show that EG‐PCD‐BiLSTM offers superior generalization ability and computational efficiency, making it robust for real‐world sentiment analysis.