Sentimental Analysis using Bi-directional RNN with Attention Mechanism based on Term Frequency- Inverse Document Frequency

Abstract

In this research, we propose a robust sentiment analysis model leveraging a Bidirectional Recurrent Neural Network (Bi-RNN) with an attention mechanism, combined with Term Frequency-Inverse Document Frequency (TF-IDF) for feature extraction. The model is designed to effectively capture the nuances of sentiment expressions in user-generated reviews, specifically focusing on the Yelp Reviews dataset. Our approach addresses the challenges of long-term dependency learning and context interpretation by employing a Bi-RNN architecture that processes text in both forward and backward directions, complemented by an attention mechanism that highlights the most relevant parts of the text. The proposed model achieved an accuracy of 93.85%, surpassing baseline models such as standard RNNs, unidirectional LSTMs, and SVM classifiers. Additionally, the model demonstrated balanced performance across positive, negative, and neutral sentiment classes, showcasing its ability to handle class imbalances and complex sentiment patterns. The research highlights the potential of the Bi-RNN with attention in enhancing sentiment analysis tasks, and suggests further improvements through the integration of advanced deep learning techniques and transformer-based architectures