Adaptive Intervention Strategies in Co-Evolving Multiplex Networks: A Reinforcement Learning Approach to Real-Time Misinformation Containment

Authors

• Anjali Ashokrao Bhadre Dr., G H Raisoni College of Engineering & Management Pune IndiaFollow
• Harshvardhan Prabhakar Ghongade Dr., Brahma Valley College of Engineering and Research InstituteFollow

Author ORCID Identifier

Anjali A. Bhadre¹* and Harshvardhan P. Ghongade²

¹ Department of Information Technology, G H Raisoni College of Engineering & Management, Pune 412207, Maharashtra, India - orcidid - 0009-0009-0734-2402

² Department of Mechanical Engineering, Brahma Valley College of Engineering and Research Institute (SPPU), Nashik, India - orcidid - 0000-0001-6840-3904

Abstract

The growing transmission of misinformation via social media creates serious challenges to public health, democracy and social cohesion. To date, methods used to contain misinformation rely upon static representations of networks and set rules for interventions. In contrast, this study presents the first Multiplex Adaptive Reinforcement Intervention Network (MARIN), a framework for real-time adaptive intervention in the context of dynamic misinformation transmission using co-evolving multiplex networks and deep reinforcement learning. Unlike past studies that have assumed static network structures, MARIN has the ability to allow for dynamic changes in network topology as a result of both misinformation transmission and intervention action. Therefore, the reinforcement learning agent is able to adapt its strategy while the structural environment continues to change. The MARIN framework uses a Graph Neural Network (GNN) encoder to encode multiplex network state and a Double Deep Q-Network (DDQN) to select optimal interventions in terms of budget constraint across three layers of the multiplex network: Information Sharing, Social Reinforcement and Private Communication. A full mathematical specification of the co-evolving network model is presented along with formal definitions of State, Action and Reward specifications. Finally, we also present the results of 100-run Monte Carlo simulations on networks with 1,000 to 5,000 nodes. These results show that MARIN obtains Misinformation Reach Reduction (MRR) improvements of 34%-47% over the best static baseline and 15%-21% over adaptive non-RL baselines under high co-evolution conditions. Additionally, we provide evidence of a critical rewiring threshold beyond which static interventions fail catastrophically and are consistent with a discontinuous phase transition exhibiting hysteresis. In conclusion, the integration of complex system science and AI provides the basis for transitioning misinformation modeling from descriptive analytical capabilities to prescriptive capabilities to enable real time intervention.

Recommended Citation

Bhadre, Anjali Ashokrao Dr. and Ghongade, Harshvardhan Prabhakar Dr. (2026) "Adaptive Intervention Strategies in Co-Evolving Multiplex Networks: A Reinforcement Learning Approach to Real-Time Misinformation Containment," Northeast Journal of Complex Systems (NEJCS): Vol. 8 : No. 2 , Article 3.
DOI: https://doi.org/10.63562/2577-8439.1155
Available at: https://orb.binghamton.edu/nejcs/vol8/iss2/3