Sentimental Manipulation on Social Media Data by Bidirectional Encoder Denotation by Transformers (BERT)

The proposed approach analyses social media sentiment using an advanced deep learning model for natural language understanding, Bidirectional Encoder Representations from Transformers (BERT). The system handles unstructured social media text using a data preprocessing pipeline, followed by BERT tokenization and fine-tuning for sentiment classification. The model is evaluated by calculating its precision, recall, accuracy, and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{F 1}$</tex>-score. The model's capability of understanding latent contextual relationships in texts is integral in detecting the sentiments, jargon, and sarcasm in social media texts. The analysis shows that BERT-based models' sentiment analysis is more accurate with 98.7 % and robust in comparison to traditional machine learning models, which clearly shows its capability for real-time performance of sentiment analysis and opinion mining from various social media platforms.