OBJECTIVES: Nowadays, social media has become the primary news source for Americans, rendering its data a valuable vehicle to examine how citizens feel about the health threats brought about by the COVID-19 pandemic. We examined risk perceptions of the public about the COVID-19 in the United States using Twitter data.

METHODS: We conducted a sentiment analysis of COVID-19 pandemic-related tweets, queried from the Twitter Historical PowerTrack between January 20, 2020, to August 27, 2020. Tweets were divided into basic units (tokens), which were passed to a sentiment classifier (Python’s scikit-learn’s logistic regression). The latter classified tweet sentiments as positive, negative, or neutral by assigning them a polarity.

RESULTS: A total of 229,388 COVID-19 pandemic related tweets were analyzed. Popular bigrams included “COVID-19,” “coronavirus,” “social distancing,” “death” and “new cases.” The polarity of the tweets was: 13.6% (30,131) positive sentiments, 85% (196,034) negative sentiments and 1.4% (3,223) neutral.

CONCLUSION: Our study demonstrates the importance and usefulness of Twitter data along with natural language processing techniques to capture an individuals’ risk perceptions about the COVID-19 in real-time. These findings can assist policymakers to better handle uncertainty in the COVID-19 response and mitigate specific public concerns about the pandemic.