JMIR Publications

ABSTRACT

Background:

The coronavirus disease 2019 (COVID-19) has continued to spread in the US and globally. Closely monitoring public engagement and perception of COVID-19 and preventive measures using social media data could provide important information for understanding the progress of current interventions and planning future programs.

Objective:

To measure the public’s behaviors and perceptions regarding COVID-19 and its daily life effects during the recent 5 months of the pandemic.

Methods:

Natural language processing (NLP) algorithms were used to identify COVID-19 related and unrelated topics in over 300 million online data sources from June 15 to November 15, 2020. Posts in the sample were geotagged, and sensitivity and specificity were both calculated to validate the classification of posts. The prevalence of discussion regarding these topics was measured over this time period and compared to daily case rates in the US.

Results:

The final sample size included 9,065,733 posts, 70% of which were sourced from the US. In October and November, discussion including mentions of COVID-19 and related health behaviors did not increase as it had from June to September, despite an increase in COVID-19 daily cases in the US beginning in October. Additionally, counter to reports from March and April, discussion was more focused on daily life topics (69%), compared with COVID-19 in general (37%) and COVID-19 public health measures (20%).

Conclusions:

There was a decline in COVID-19-related social media discussion sourced mainly from the US, even as COVID-19 cases in the US have increased to the highest rate since the beginning of the pandemic. Targeted public health messaging may be needed to ensure engagement in public health prevention measures until a vaccine is widely available to the public.