Who is mentioning COVID-19 articles on twitter? Classifying twitter users in the context of scholarly communication

Abstract

This study aims to examine the demographics of participants engaged in scholarly communication on Twitter, which has been rebranded as X. Firstly, based on a dataset of tweets citing COVID-19 publications, it proposed a more precise classification system consisting of eleven user categories for individuals who tweeted academic publication. Secondly, it explores the effectiveness of graph neural network models (GNNs) in combination with a transformer-based text classification model (specifically, BERT) to classify these newly defined user categories. The findings of this research highlight that GNNs can effectively interpret the social networks within scholarly communication, and complement text classification models in characterizing user types. The best-performing model achieved an accuracy rate of 84.05 percent in classifying user categories for a dataset of 10,048 labeled users. Subsequently, this model was employed to analyze 393,030 tweeters in our dataset. The analysis revealed that relevant scholarly discussion on Twitter was dominated by members from the general public (over 71 percent). Academic researchers and institutions constituted 12.48 percent, while health science professionals and institutions made up 7.35 percent of the contributors to relevant scholarly discussions on Twitter. Notably, academic publishers and research feed accounts exhibited aggressive tweeting behaviors and were responsible for the highest volume of tweets on average. This study also demonstrates the active involvement of various non-academic members, including commercial businesses, mass media outlets, public authorities, politicians, and civil society organizations, in Twitter scholarly communication.