Comparative Study of Deep Learning Models for Sentiment Analysis on Social Media Data Using Hybrid BiLSTM-CNN Architecture
Yatendra Sahu, Vivek Sharma, Arpita Bhargava, Ghanshyam Singh Thakur, Faraz Siddiqui, Bhupendra Singh Kirar · 2025
In the modern era, social media platforms serve as key venues where individuals engage in conversations, share ideas, and express opinions. Sentiment analysis is a subfield of Natural Language Processing that seeks to examine public emotions and opinions, offering valuable insights helping businesses, governments and organisations to connect with public sentiments. This work investigates Deep learning models such as Convolutional Neural Networks (CNNs), Long Short-Term Memory Networks (LSTMs), Single and Multilayer Bidirectional LSTMs (BiLSTMs), and Hybrid Neural Networks. Using the Sentiment140 dataset, preprocessing techniques were utilised to clean noisy data, and GloVe embedding were applied for improved word representation. The hybrid BiLSTM-CNN model successfully combined CNN’s capacity to extract local semantic patterns with BiLSTM’s strength in capturing contextual features, outperforming all other models with an accuracy of 76.54%.