Context-based sentiment analysis using a BiGRU DistilBERT fusion model for COVID-19 tweets

Area of Science:

• Computational Social Science
• Natural Language Processing
• Public Health Informatics

Background:

• The COVID-19 pandemic generated extensive public discourse on social media platforms like Twitter.
• Understanding public sentiment during health crises is crucial for effective policymaking and crisis management.

Purpose of the Study:

• To develop and evaluate a hybrid deep learning model for accurate sentiment analysis of COVID-19-related tweets.
• To investigate the temporal evolution and regional variations of public sentiment during the early stages of the pandemic.

Main Methods:

• A fusion model combining Bidirectional Gated Recurrent Unit (BiGRU) and DistilBERT transformer was developed.
• Features from BiGRU and DistilBERT were concatenated and fed into an XGBoost meta-classifier.
• The model was trained and validated on over one million English-language tweets from eight countries (January-April 2020).

Main Results:

• The fusion model achieved 85.8% classification accuracy, outperforming individual models like DistilBERT (85.5%).
• Public sentiment mirrored pandemic phases, with negative sentiment peaking during case/death surges and positive sentiment during recovery.
• Significant regional differences in sentiment trends were observed across countries.

Conclusions:

• The proposed context-infused fusion model effectively captures public sentiment dynamics during health crises.
• Findings provide actionable insights for policymakers regarding social media monitoring and public communication strategies.
• The study highlights the importance of considering temporal and regional factors in sentiment analysis.