Applying and Understanding an Advanced, Novel Deep Learning Approach: A Covid 19, Text Based, Emotions Analysis Study

Area of Science:

• Computational Social Science
• Natural Language Processing (NLP)
• Artificial Intelligence (AI)

Background:

• The COVID-19 pandemic significantly impacted global emotional well-being, necessitating an understanding of public sentiment.
• Microblogging platforms like Twitter became crucial for individuals to express feelings and concerns during the pandemic.
• Limited research existed on expressed emotional well-being during the early stages of the COVID-19 pandemic.

Purpose of the Study:

• To globally identify, explore, and understand emotions expressed on Twitter during the initial months of the COVID-19 pandemic.
• To utilize Deep Learning and Natural Language Processing (NLP) for analyzing public sentiment and emotional variations.

Main Methods:

• Collected and analyzed over 2 million tweets from February to June 2020.
• Employed Transfer Learning and the Robustly Optimized BERT Pretraining Approach (RoBERTa) for deep learning analysis.
• Utilized a Reddit-based standard Emotion Dataset for transfer learning and developed a multi-class emotion classifier.

Main Results:

• Achieved a tweet classification accuracy of 80.33% and an average MCC score of 0.78, outperforming existing AI methods.
• Demonstrated the novel application of the RoBERTa model for analyzing public emotional well-being during the pandemic.
• Provided insights into the temporal changes in citizens' emotional well-being worldwide.

Conclusions:

• The study offers valuable data mining and analytics insights for the pandemic era.
• Findings can inform effective pandemic management strategies and predictive models for national emotional well-being.
• Highlights the potential of AI and NLP in understanding societal responses to global crises.