Enhancing sentiment analysis accuracy on social media comments using a tuned BERT model
Utpol Kanti Das, R. Ani, Nippon Datta, Iftikharul Fahad, Juel Sikder, Umme Sara, Arpita Chakraborty
Abstract
Sentiment analysis of social media comments is essential for real-time comprehension of user input and public opinion. To facilitate finding relevant news and views on social networks, this study proposes a model that categorizes search results by classifying sentiments in social comments and news articles. The research explores how a tuned BERT (Bidirectional Encoder Representations from Transformers) model can enhance sentiment analysis accuracy. The process begins with preprocessing comments to remove unnecessary elements, followed by feature extraction and text classification using trained models. The study compares the tuned BERT model’s performance against established models like Support Vector Machines (SVM), Naive Bayes, Long Short-Term Memory Networks (LSTM), and Convolutional Neural Networks (CNN). Using the Multi-Class Sentiment Analysis Dataset (MCSAD), the models are evaluated on accuracy, precision, recall, and F1-score metrics. The results demonstrate that the tuned BERT model outperforms traditional models, achieving 99.60% accuracy without augmentation and 99.98% accuracy with data augmentation, especially in handling the complex language of social media. The study also highlights the strengths and weaknesses of different models, offering insights for future research and practical applications in sentiment analysis.