Abstract:
The COVID-19 pandemic has led to a misinformation avalanche on social media, which produced confusion and insecurity in netizens. Learning how to automatically recognize adoption or rejection of misinformation about COVID-19 enables the understanding of the effects of exposure to misinformation and the threats it presents. By casting the problem of recognizing misinformation adoption or rejection as stance classification, we have designed a neural language processing system operating on micro-blogs which takes advantage of Graph Attention Networks relying on lexical, emotion, and semantic knowledge to discern the stance of each micro-blog with respect to COVID-19 misinformation. This enabled us not only to obtain promising results, but also allowed us to use a taxonomy of COVID-19 misinformation themes and concerns to characterize the misinformation adoption or rejection that can be best recognized automatically.
