Polarity Related Influence Maximization through Multi-Agent Reinforcement Learning (2026)

Proceedings of the 18th International Conference on Agents and Artificial Intelligence

Authors

Anikó Kopacz, Camelia Chira

Abstract

Influence maximization is a network optimization problem, which consists of selecting nodes as sources while maximizing the spread of information. The source nodes that are initially activated form the seed set. Polarity-related influence maximization accounts for having both positive and negative types of relationships between nodes. In this paper, we propose a two-stage multi-agent reinforcement learning based approach to address the polarity-related influence maximization problem. In the first stage, Louvain community detection is applied to find interlinked groups of nodes, and we assign agents to a subset of the communities. The responsibility of the agents is to select a strategy to determine a node from the community as the next seed-node. In the second stage, a deep reinforcement learning model is trained to select the strategies that maximize the information spread in the network. The nodes determined by the agents are aggregated and form the seed set. The proposed approach is validated on the Bitcoin OTC, WikiElec and Slashdot directed signed networks, and the results show that community-based reinforcement learning agents are able to optimize the positive influence spread.

Citation

@Inproceedings{Kopacz2026PolarityRI,
 author = {Anikó Kopacz and Camelia Chira},
 booktitle = {Proceedings of the 18th International Conference on Agents and Artificial Intelligence},
 title = {Polarity Related Influence Maximization through Multi-Agent Reinforcement Learning},
 year = {2026}
}