A Multiagent Reinforcement Learning algorithm to solve the Community Detection Problem

Community detection is a challenging optimization problem that consists of searching for communities that belong to a network under the assumption that the nodes of the same community share properties that enable the detection of new characteristics or functional relationships in the network. Although there are many algorithms developed for community detection, most of them are unsuitable when dealing with large networks due to their computational cost. In this paper, a new multiagent reinforcement learning algorithm is proposed for community detection in complex networks. Each agent in the multiagent system is an autonomous entity with different learning parameters and based on the cooperation among the learning agents and updating action probabilities of each agent, the algorithm interactively will identifies a set of communities in the input network that are more densely connected than other communities. In other words, a number of independent agents interactively attempt to identify communities and evaluate the quality of the communities found at each stage by the normalized cut as objective function and update their probability vectors based on the results of the evaluation. If the quality of the community found by an agent in each of the stages, better than all the results produced so far, is referred as the successful agent and the other agents will update their probability vectors based on the result of the successful agent.  In the experiments, the performance of the proposed algorithm is validated on four real-world benchmark networks and twelve synthetic LFR networks. Experimental results in comparison with four state-of-the-art community detection algorithms in terms of modularity and NMI show that our algorithm outperforms these algorithms and can detect communities in small-scale and large networks with high speed, accuracy and stability and is capable of managing large-scale networks up to 5000 nodes.