Exploratory community detection: Finding communities in unknown networks

Community detection amounts to one of the key methods in handling social networks with the aim of capturing global patterns of a network. This paper focuses on a situation where the network is unknown, which would render existing algorithms unusable. We propose exploratory community detection which aims to detect communities by utilizing samples taken from diffusion process over the network. For this problem, we propose a neural-based algorithm that develops a matrix representation of the network structure. This matrix is then the input of a spectral clustering algorithm to reveal communities in the network. We perform experiments on real-world and synthetic data sets with simulated diffusion samples.The results reveal that our algorithm has strong empirical performance.