AI-Powered Sentiment Analysis

Hybrid Deep Learning Architecture

Our proposed fusion model combines the strengths of various deep learning models with a classical meta-classifier for robust sentiment prediction. The architecture integrates:

• Bidirectional GRU (BiGRU): Processes text in both forward and backward directions to capture long-range dependencies and dual context.
• DistilBERT: A powerful transformer model, pre-trained on vast English corpuses, providing efficient and accurate contextual embeddings.
• FastText: Utilizes character n-grams to handle rare words and misspellings, crucial for noisy social media data.

The outputs from these base learners (BiGRU, FastText, DistilBERT) are concatenated into a comprehensive feature vector, which is then fed into an XGBoost meta-classifier. XGBoost effectively learns to weight and integrate the base models' predictions, resulting in a highly accurate final sentiment label. This stacked generalization approach enhances overall classification accuracy and confidence.

Rigorous Performance Evaluation

Our model was trained and validated on the Stanford Sentiment140 dataset, comprising 1.6 million English tweets. The fusion model (BiGRU + DistilBERT + XGBoost) achieved a remarkable classification accuracy of 85.8%, outperforming individual models and other traditional deep learning approaches.

Key performance metrics consistently demonstrate the superiority of our hybrid approach across accuracy, F1-Score, precision, recall, sensitivity, and specificity. This robust performance validates the model's ability to handle the complexities and inherent noise of real-world social media data during a crisis, ensuring reliable sentiment detection.

Actionable Insights for Policymakers

The findings from our context-based sentiment analysis offer valuable, actionable insights for public health authorities and policymakers during health crises. By correlating tweet sentiment trends with real-world events like case surges and recovery periods, we can:

• Track Public Anxiety: Identify peaks in negative sentiment coinciding with rising cases and deaths.
• Gauge Optimism: Observe positive sentiment aligning with recovery phases.
• Understand Regional Variations: Analyze how sentiment differs across countries due to socio-economic, cultural, and political factors, allowing for tailored communication strategies.
• Combat Misinformation: By understanding the prevailing sentiment, authorities can proactively address public concerns and improve public morale.

This dynamic understanding of public sentiment enables more effective communication strategies and targeted interventions during critical times, ultimately fostering better public responses and outcomes.