Monte Carlo Tree Search for Community Detection MCTS-CD
Monte Carlo Tree Search pour la détéction de communautés MCTS-CD
Résumé
Real-world graph analysis and the study of complex networks encounter the common challenge of community detection(CD), encompassing diverse domains such as transportation networks, cyber-security, animal, and social media networks. This problem can be modeled as an NP-hard optimization problem. Inspired from [25] [24] that combined Monte Carlo Tree Search (MCTS) and Reinforcement Learning on large action-space environments, we adapt MCTS for the problem of Community Detection in Social Networks and propose MCTS-CD. MCTS-CD is an arm-bandit [2] anytime algorithm. We show that when compared to reference algorithms for CD such as Louvain [1], with allocating the same amount of resources, MCTS-CD acheives better results on many benchmark datasets such as Zachary Karate Club, Dolphin and the Pol Books datasets.
| Origine | Fichiers produits par l'(les) auteur(s) |
|---|
