An authorship-centric AI framework for forensic identification of coordinated actors in social media astroturfing campaigns
- 1Symbiosis Institute of Computer Studies and Research (SICSR), Symbiosis International (Deemed University)
Abstract
Astroturfing on social media has become increasingly complex and often continues even after it has been identified as coordinated inauthentic activity. Although existing studies mainly focus on detecting astroturfing, they provide limited insight into how such campaigns are carried out—specifically, whether the coordination is driven by a single individual, a group of coordinated humans, or automated bot networks. To address this gap, the AstroAuth-ID framework is proposed to analyse coordinated astroturfing activity in social media datasets through an authorship-centric, multi-view learning approach. The framework constructs user-level author profiles by combining semantic representations obtained from large language models with writing-style features, posting-time patterns, and interaction-based behavioural signals. These diverse features are integrated into a single representation and further enhanced using graph-based representation learning to model relationships and coordination among accounts. Unsupervised and density-based clustering techniques, including Gaussian Mixture Models (GMM) and HDBSCAN, are applied to identify latent coordination structures to determine the dominant coordination type. The proposed AstroAuth-ID framework demonstrates strong coordination inference capabilities across various social media datasets because it achieved DBCV scores that reached 0.78 and Dunn Index values that exceeded 3.36, while maintaining low S_Dbw scores and controlling cluster entropy to demonstrate strong cluster compactness, separation and structural coherence. Experimental results show that the framework consistently captures diverse coordination dynamics through integrated behavioural, semantic, stylometric, interaction, and graph-based representations. The framework effectively distinguishes bot-driven campaigns, human-operated sock puppet networks, and large-scale hybrid (cyborg) influence operations, even in sparse and adversarial environments. Overall, AstroAuth-ID provides accurate astroturfing detection along with interpretable forensic insights into coordinated influence structures and operational behaviour.
Related
- Gender Disparity in Movement Disorders Research Authorship: Progress and Persistent Gaps Across a Decade
- Trends in diversity of scientific authorship by ethnicity, sex and age of authors in a National Institute for Health and Care Research Biomedical Research Centre in England: a retrospective longitudinal bibliometric study.
- FedMann: A Manually Coordinated Federated Learning Framework Applied to Multiple Myeloma—A Feasibility Study
- Ensuring quality in research, 'Authorship Effect' & Conflict of Interest. Response submitted to BMJ Mental Health in response to 'Clinician perspectives on the utility and acceptability of OxSATS: a qualitative study of a novel structured suicide risk assessment tool
- Identification and validation of feature genes using a ferroptosis-informed analytical framework in medulloblastoma
BibTeX
@article{badade2026authorshipcentric,
title = {{An authorship-centric AI framework for forensic identification of coordinated actors in social media astroturfing campaigns}},
author = {Akshata Balasaheb Badade and Rajesh Kumar Dhanaraj},
journal = {EPJ Data Science},
year = {2026},
month = jul,
doi = {10.1140/epjds/s13688-026-00683-w},
url = {https://doi.org/10.1140/epjds/s13688-026-00683-w},
note = {CC BY},
}
RIS
TY - JOUR
TI - An authorship-centric AI framework for forensic identification of coordinated actors in social media astroturfing campaigns
AU - Akshata Balasaheb Badade
AU - Rajesh Kumar Dhanaraj
PY - 2026
DA - 2026/07/01
JO - EPJ Data Science
DO - 10.1140/epjds/s13688-026-00683-w
UR - https://doi.org/10.1140/epjds/s13688-026-00683-w
AB - Astroturfing on social media has become increasingly complex and often continues even after it has been identified as coordinated inauthentic activity. Although existing studies mainly focus on detecting astroturfing, they provide limited insight into how such campaigns are carried out—specifically, whether the coordination is driven by a single individual, a group of coordinated humans, or automated bot networks. To address this gap, the AstroAuth-ID framework is proposed to analyse coordinated astroturfing activity in social media datasets through an authorship-centric, multi-view learning approach. The framework constructs user-level author profiles by combining semantic representations obtained from large language models with writing-style features, posting-time patterns, and interaction-based behavioural signals. These diverse features are integrated into a single representation and further enhanced using graph-based representation learning to model relationships and coordination among accounts. Unsupervised and density-based clustering techniques, including Gaussian Mixture Models (GMM) and HDBSCAN, are applied to identify latent coordination structures to determine the dominant coordination type. The proposed AstroAuth-ID framework demonstrates strong coordination inference capabilities across various social media datasets because it achieved DBCV scores that reached 0.78 and Dunn Index values that exceeded 3.36, while maintaining low S_Dbw scores and controlling cluster entropy to demonstrate strong cluster compactness, separation and structural coherence. Experimental results show that the framework consistently captures diverse coordination dynamics through integrated behavioural, semantic, stylometric, interaction, and graph-based representations. The framework effectively distinguishes bot-driven campaigns, human-operated sock puppet networks, and large-scale hybrid (cyborg) influence operations, even in sparse and adversarial environments. Overall, AstroAuth-ID provides accurate astroturfing detection along with interpretable forensic insights into coordinated influence structures and operational behaviour.
SN - 2193-1127
ER -