Reading Emotions in the Digital Age: A Deep Learning Approach to Detecting Anxiety During the COVID-19 Pandemic Through Social Media
During an unprecedented crisis, such as the COVID-19 pandemic, people respond to increased uncertainty in their social surroundings, which often entails various emotional expressions, including fear and frustration. Anxiety is particularly important because it may serve as a symptomatic indicator of various social problems, such as the collapse of trust, polarized public opinions, and an increase in violence. Identifying the expressions of anxiety and tracking their fluctuations over time may lead to a better understanding of how people collectively cope with social crises. This study aims to develop a deep learning-based classification of anxiety and track how the degree of anxiety changed over time in the context of the COVID-19 pandemic. Using the bidirectional encoder representations from transformers (BERT) model, we extracted anxiety-laden messages from Twitter and examined how the longitudinal distribution of anxiety corresponded to the major waves of COVID-19 using intervention time-series analyses.
social media, anxiety, COVID-19, deep learning, time-series analysis