A Combinatorial Multi-Armed Bandit Based Method for Dynamic Consensus Community Detection in Temporal Networks

Community detection in temporal networks is an active field of research, which can be leveraged for several strategic decisions, including enhanced group-recommendation, user behavior prediction, and evolution of user interaction patterns in relation to real-world events. Recent research has shown that combinatorial multi-armed bandit (CMAB) is a suitable methodology to address the problem of dynamic consensus community detection (DCCD), i.e., to compute a single community structure that is conceived to be representative of the knowledge available from community structures observed at the different time steps. In this paper, we propose a CMAB-based method, called CreDENCE, to solve the DCCD problem. Unlike existing approaches, our algorithm is designed to provide a solution, i.e., dynamic consensus community structure, that embeds both long-term changes in the community formation and newly observed community structures. Experimental evaluation based on publicly available real-world and ground-truth-oriented synthetic networks, with different structure and evolution rate, has confirmed the meaningfulness and key benefits of the proposed method, also against competitors based on evolutionary or consensus approaches.