Disentangling User Types in Twitter: A Profile Classification Model Development

• In Event: Computational methods for Analyzing Visual and Online Communication

Abstract

The prevalence of large-scale social media research on human behaviors has raised concerns regarding the ecological validity of social media data, especially whether it accurately represents the population it claims to study. Communication research is not immune to sampling bias. The current project highlights one specific sampling issue known as proxy population mismatch. This project develops a computational model that separates ordinary user accounts, whom communication scholars traditionally envision as members of “the public”, from non-laymen accounts (e.g. non-personal actors, experts, politicians, media personnel) in Twitter. The process of machine-learning model development, quality of the model outcomes, and application to terrorism news Twitter data are discussed.

Authors

• K. Hazel Kwon, Arizona State U

• Hunter Priniski, Arizona State U

• Monica Chadha, Arizona State U