Enhancing Sentiment Analysis Using CNN: Evaluative Study of Traditional Models

Abstract

This paper forms a research work in which a novel method for performing sentiment analysis is introduced in the form of convolutional neural networks and tested against other established algorithms including logistic regression, Naive Bayes, and decision tree classifiers. Our proposed method is to apply deep learning approaches to classify appropriate sentiments associated with the textual data, which are crucial for estimating users’ satisfaction, their mood changes depending on the environment, or how they respond to specific events. CNNs have shown their capability in image analysis through the feature of being able to extract and analyze local features with robustness. This capability is extended to text data as well while at the same time preserving the context within which the sentiment of the data is being expressed. For multiple datasets, it has been observed that accuracy and F1 score were higher for the proposed custom CNN model as compared to traditional methods. As confirmed by these results, CNNs can contribute to raising the sentiment analysis and sentiment prediction accuracy in the different application contexts.