A Compression-Based Approach to Detecting Automated and Coordinated Behavior on Social Media

Social media platforms are frequently targeted by entities engaging in automated or coordinated behavior, aiming to manipulate public opinion or conduct information operations without revealing their synthetic or managed nature. Research on detecting such actors faces the challenge of developing scalable, versatile methods that allow for consistent comparisons across diverse datasets. The challenge is even made more pressing by evidence of these actors on platforms beyond the extensively studied X (formerly Twitter), as well as the emergence of new platforms. We fill this gap by introducing a novel compression-based detection methodology, in addition to a new sparse method for network reconstruction that scales linearly under reasonable parameter choice. Being independent of the social media platform and the behavioral trace under study, our approach marks a departure from traditional methods that rely on multiple criteria or measures to assess user similarity. We evaluate our technique on multiple benchmark and real-world datasets, including widely known datasets related to political campaigns and emerging misinformation scenarios. We show that our approach provides a flexible unsupervised framework that effectively identifies both automated and coordinated activities across various behavioral traces, ensuring broad applicability.