Understanding conflict origin and dynamics on Twitter: A real-time detection system

• Published in: Expert systems with applications Vol. 212; p. 118748
• Main Authors: de França, Fabricio Olivetti, di Genova, Daniel Vitor Beraldo, Penteado, Claudio Luis Camargo, Kamienski, Carlos Alberto
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.02.2023
• Subjects: Conflicts dynamics; Explainable AI; Social Network Analysis; Social Network Analysis; Explainable AI; Conflicts dynamics
• ISSN: 0957-4174; 1873-6793
• DOI: 10.1016/j.eswa.2022.118748

Social Network Sites provide a venue for people worldwide to share their point of view and interact with each other, offering a virtual space with freedom for expressing ideas and opinions. The interaction dynamics often creates clusters of users sharing similar interest and opinions, thus creating an information bubble or echo chamber. In certain topics, such as politics, different groups tend to collide and start arguments characterized by conflicts of opinion. This fact has been increasingly observed during the COVID-19 pandemic, fed by misinformation and anti-science movements. One approach to address these issues is to use statistical measures of the posts revolving around the topic of interest, such as the number of shares, likes, and replies. In this paper we propose a methodology to extract a feature set from trending topics of the Twitter social network and apply two white-box models, a Symbolic Regression, named ITEA, and a Decision Tree, for the automated detection and understanding of conflicts. Our experiments show that both models obtain close extrapolation accuracy to the baseline black-box model (Random Forest). As a highlight of this paper, both white-box models are fully described to be used by any practitioner. Additionally, the model created by ITEA allows us to extract some insights from the generated models. Although these models do not allow for a complete comprehension of the dynamics of a conflict, it certainly points towards a direction for a more thorough investigation. •Classifiers to detect conflicts of opinion on Twitter using statistical metrics.•Use of transparent models capable of helping to understand the dynamics of conflicts.•Analysis of the conflict dynamics using the generated transparent models.