Thu 4 Feb 21
A team of data scientists including Essex academics has analysed 94 million tweets from the first months of the pandemic to track COVID-19’s effect on mental health.
The research team used machine learning to develop a model able to capture data indicating depression, stress, anxiety and suicidal thoughts among users of the social media platform. The aim was to tap into popular technology to help public health experts identify changes in community levels of depression over time.
The World Health Organisation highlighted early in 2020 that the pandemic would likely have a negative impact on mental health, with the disease affecting many facets of life including work, health and relationships.
Researchers from the University of Technology Sydney (UTS) and the University of Essex developed their novel classification model to tease out the psychological impact of COVID-19 outbreaks and government policies such as lockdowns on Australian users.
While Australia has been less affected by COVID-19 than other countries around the globe, the results show the first wave of cases and the resulting lockdown still had a profound impact on mental health in the community.
“Twitter is used by millions of users throughout the globe where they share their thoughts, their lives, places they visit, photos, among various other things,” explained Dr Shoaib Jameel, from Essex’s School of Computer Science and Electronic Engineering. “During the current pandemic, we have been confined within the borders of our homes without much social contact. As a result, people are now using Twitter to share their feeling which might include depression-related cues without many people explicitly mentioning that they are depressed.”
He added: “This study is important because it helps us understand the timelines when people went through depression, such as lockdown versus no lockdown. Whilst the study has been conducted on users in Australia, the findings can be generalised to any region.”
Professor Guandong Xu, from the UTS School of Computer Science, added: “Identifying community depression dynamics can help governments and policymakers better understand the psychological impacts of policy decisions, and identify communities that may require increased public health support.”
The researchers captured and analysed data from 94 million tweets posted by social media users in New South Wales between 1 January and 22 May 2020.
The machine-learning based depression detection model classified the content of tweets according to topic, emotion – including the use of emojis – and recognised symptoms of depression such as fatigue, weight loss, feelings of worthlessness and suicidal ideation.
The model revealed a significant jump in depression levels at the start of the COVID-19 outbreak in New South Wales, reaching a peak on 26 March 2020 that coincided with the highest number of recorded cases.
Government measures such as the state lockdown on 31 March appeared to slightly increase depression levels, although easing lockdown did not reduce depression.
The researchers also found that during lockdown more than 40% of Twitter users increased the time they spent on the platform.
The paper, Detecting Community Depression Dynamics Due to COVID-19 Pandemic in Australia, is published in IEEE Transactions on Computational Social Systems.