Classification and Keyword Identification of COVID-19 Misinformation on Social Media: A Framework for Semantic Analysis
Abstract
The growing surge of misinformation among COVID-19 communication can pose great hindrance to truth, magnify distrust in policy makers and/or degrade authorities' credibility, and it can even harm public health. Classification of textual context on social media data relating to COVID-19 is an effective tool to combat misinformation on social media platforms. In this research, Twitter data was leveraged to (1) develop classification methods to detect misinformation and identify Tweet sentiment with respect to COVID-19 and (2) develop a human-in-the-loop interactive framework to enable identification of keywords associated with social context, here, being misinformation regarding COVID-19. (1) Six fusion-based classification models were built fusing three classical machine learning algorithms. The best performing models were selected to detect misinformation and to classify sentiment. We found the public reacted more positively towards COVID-19 misinformation and positive sentiment increased in August 2020 relative to April 2020 for all but political or biased related misinformation. (2) The most semantically similar keywords were chosen via distribution representations of topics and recommended by optimal ROC curves. The interactive framework recommended 21 and 22 keywords related to conspiracy and unreliable misinformation, respectively, and are most semantically similar to the user inquiry "COVID start lab."
Document Details
- Document Type
- Technical Report
- Publication Date
- Mar 01, 2022
- Accession Number
- AD1166828
Entities
People
- Grace Y. Smith
Organizations
- Air Force Institute of Technology