Can User Characteristics Predict Norm Adherence on Social Media? Exploring User-Centric Misinformation Interventions

Abstract

The work-in-progress explores user-centric misinformation interventions on social media as such tools are limited. Knowledge of relationships between online user characteristics and their expressed adherence to a desired norm (i.e., rejecting misinformation or supporting factual information) is understudied with limited integrated multi-modal machine-learning models to infer demographic and sociopsychological characteristics. Thus, we piloted 9,331 Twitter users tweeting COVID-19 vaccines between May 1, 2020, and April 30, 2021. Employing a CNN-LSTM framework, our model analyzes user biographies, profile images, and pre-COVID historical tweets to infer user traits over 90% accuracy for individual characteristics and an overall accuracy of 85.61%, which outperforms existing tools and other designs. Further, using multi-logistic regression, we identified significant predictors of users' adherence to desired norms, such as gender and pre-pandemic prosocial content engagement, while finding no significant age correlation. Our findings illuminate pathways for targeted misinformation mitigation strategies during critical public health crises.