Using graph embedding and machine learning to identify rebels on twitter

• Published in: Journal of informetrics Vol. 15; no. 1; p. 101121
• Main Authors: Masood, Muhammad Ali, Abbasi, Rabeeh Ayaz
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.02.2021
• Subjects: Machine learning; Rebels; Social network analysis; Supervised Rebel Identification (SRI); User graph; User graph; Social network analysis; Supervised Rebel Identification (SRI); Rebels; Machine learning
• ISSN: 1751-1577; 1875-5879
• DOI: 10.1016/j.joi.2020.101121

•A new multi-regional dataset having tweets of rebel, normal and counter rebel users covering five countries is presented•A novel user graph to capture stance of a user is proposed•Fifteen features belonging to three categories are proposed•Various aspects of normal, rebel and counter rebel users are analyzed During the last two decades, the number of incidents from extremists have increased, so as the use of social media. Research suggests that extremists use social media for reaching their purposes like recruitment, fund raising, and propaganda. Limited research is available to identify rebel users on social media platforms. Therefore, we propose a Supervised Rebel Identification (SRI) framework to identify rebels on Twitter. The framework consists of a novel mechanism to structure the users’ tweets into a directed user graph. This user graph links predicates (verbs) with the subject and object words to understand semantics of the underlying data. We convert the user graph into graph embedding to use these semantics within the machine learning algorithms. Apart from the user graph and its embedding, we propose fourteen other features belonging to tweets’ contents and users’ profiles. For evaluation, we present the first multicultural and multiregional dataset of rebels affiliated with nine rebel movements belonging to five countries. We evaluate the proposed SRI framework against two state-of-the-art baselines. The results show that the SRI framework outperforms the baselines with high accuracy.