Optimism and pessimism analysis using deep learning on COVID-19 related twitter conversations

Area of Science:

• Natural Language Processing
• Computational Social Science
• Artificial Intelligence

Background:

• The COVID-19 pandemic generated vast amounts of social media data, reflecting public sentiment.
• Understanding emotional dynamics in online health crisis conversations is crucial for public health response.

Purpose of the Study:

• To develop and evaluate a deep learning approach for detecting optimism and pessimism in Twitter conversations related to COVID-19.
• To analyze emotional shifts and their relationship with social influence in crisis communication.

Main Methods:

• Utilized a pre-trained transformer embedding for semantic feature extraction.
• Compared various network architectures, with Bidirectional Long- and Short-Term Memory (BiLSTM) networks showing superior performance.
• Evaluated models on two novel, publicly available Twitter corpora comprising 150,503 tweets from 51,319 users.

Main Results:

• The best models accurately detected optimism and pessimism across different pandemic phases.
• Conversations with strong pessimistic signals exhibited minimal emotional shifts (62.21%).
• Optimistic conversations were more likely to maintain their mood (only 10.42% maintained mood).
• User emotional volatility correlated with social influence.

Conclusions:

• The proposed deep learning approach effectively models optimism and pessimism in health crisis contexts.
• Emotional signals and shifts in Twitter conversations reveal empathy and support mechanisms.
• Findings highlight the distinct emotional trajectories of optimistic versus pessimistic online discussions during a pandemic.