Abstract
In this study we investigate the characteristics of malicious account behaviors on Twitter based on the analysis of the published data archive. We investigate emergent behavior of malicious accounts that Twitter tagged as connected to state-backed information operations, identified as malicious and removed from the Twitter network. We focus on the analysis of four types of malicious accounts’ features: (1) Account reputation, (2) Account tweeting frequency, (3) Age of account and (4) Account activity score. With the use of descriptive statistics and unsupervised learning, we attempt to extend past research that defined behavioral patterns of malicious actors on Twitter. Our research contributes to the understanding of behavior of malicious actors and enriches current research in that area. In this paper we analyze the dataset published by Twitter in January 2019, which contains details on suspended malicious accounts’ activities initiated in Bangladesh. To the best of our knowledge, this article is the first effort to extend research on malicious account behavior based on labeled proprietary data on removed malicious accounts identified and released by Twitter itself.
Recommended Citation
Onuchowska, Agnieszka and Berndt, Donald J., "Towards Understanding Malicious Actions on Twitter" (2019). AMCIS 2019 Proceedings. 2.
https://aisel.aisnet.org/amcis2019/rhetoric_social_media_disinformation/rhetoric_social_media_disinformation/2
Towards Understanding Malicious Actions on Twitter
In this study we investigate the characteristics of malicious account behaviors on Twitter based on the analysis of the published data archive. We investigate emergent behavior of malicious accounts that Twitter tagged as connected to state-backed information operations, identified as malicious and removed from the Twitter network. We focus on the analysis of four types of malicious accounts’ features: (1) Account reputation, (2) Account tweeting frequency, (3) Age of account and (4) Account activity score. With the use of descriptive statistics and unsupervised learning, we attempt to extend past research that defined behavioral patterns of malicious actors on Twitter. Our research contributes to the understanding of behavior of malicious actors and enriches current research in that area. In this paper we analyze the dataset published by Twitter in January 2019, which contains details on suspended malicious accounts’ activities initiated in Bangladesh. To the best of our knowledge, this article is the first effort to extend research on malicious account behavior based on labeled proprietary data on removed malicious accounts identified and released by Twitter itself.