Classification of social media users with generalized functional data analysis

Abstract

Technological advancement has made possible the collection of data from social media platforms at unprecedented speed and volume. Current methods for analyzing such data lack interpretability, are computationally intensive, or require a rigid data specification. Functional data analysis enables the development of a flexible, yet interpretable, modeling framework to extract lower-dimensional relevant features of a user's posting behavior on social media, based on their posting activity over time. The extracted features can then be used to discriminate a malicious user from a genuine one. The proposed methodology can classify a binary time series in a computationally efficient manner and provides more insights into the posting behavior of social media agents. Performance of the method is illustrated numerically in simulation studies and on a motivating Twitter data set. The developed methods are applicable to other social media data, such as Facebook, Instagram, Reddit, or TikTok, or any form of digital interaction where the user's posting behavior is indicative of their user class.