JMIR Publications

ABSTRACT

Background:

During rapidly developing public health crises, such as the COVID-19 pandemic, continuous monitoring of the psychological well-being of the population is crucial. However, the real-time surveillance of the population’s psychological health remains challenging, mainly due to the lack of suitable data and techniques that can be used for monitoring purposes.

Objective:

This study proposes a machine learning framework for the real-time surveillance of mental health conditions that does not require extensive training data. Using survey-linked tweets, we then track the level of emotional distress during the COVID-19 pandemic separately by attribute and psychological condition of social media users in Japan.

Methods:

We conducted online surveys on the general adult population in Japan in May 2022 and collected their basic demographic information, socioeconomic status, and mental health conditions along with their Twitter handle (N=2432). After verifying the validity of their account, we retrospectively collected all the tweets posted by the study participants between January 1, 2019, and May 30, 2022 (N= 2,493,682). After excluding invalid posts and users, our dataset contained 495,021 tweets generated by 560 individuals (ages 18-49) in 2019 and 2020. We computed emotional distress scores for each tweet using a semi-supervised algorithm called Latent Semantic Scaling, with higher values indicating higher levels of emotional distress. We estimated fixed-effect regression models to examine if the emotional distress levels in 2020 were different from the corresponding weeks in 2019, while controlling for the underlying emotional distress levels of individuals.

Results:

The level of emotional distress increased in the week when the school closure started (March 2020), and it peaked at the beginning of the state of emergency period (0.219, 95% CI: 0.162–0.276) in early April 2020. The level of emotional distress was unrelated to the number of COVID-19 cases. We found that the government-induced restrictions in the early phase of the COVID-19 pandemic disproportionately affected the psychological conditions of vulnerable individuals, including those with low income, precarious employment, depressive symptoms, and suicidal ideation. At the same time, we did not find any evidence that the media reporting on celebrities’ suicide affected the emotional distress levels of Twitter users, including those with depressive symptoms and suicidal ideation.

Conclusions:

Our results suggest that the containment policies, not the spread of the disease itself or associated fears, were a source of emotional distress for many individuals in Japan during the early months of the pandemic. The present study establishes a framework to implement near real-time monitoring of the emotional distress level of social media users, highlighting a great potential to continuously monitor the psychological health of the population using social media data as a complement to administrative and large-scale survey data.