Title: Automatic detection of multilevel communities: scalable and resolution-limit-free

Abstract: Community detection in complex networks has been hindered by two defects: (1) the resolution limit problem, which restrains simultaneous detection for communities of heterogeneous sizes, and (2) divergent outputs of the optimization algorithm, which set hurdles for the differentiation of more relevant and significant results. In this paper, we suggest a renewed method for community detection via a scalable community fitness function. Due to its scalability, this method is on the one hand free of the resolution limit problem, even in large heterogeneous networks, and on the other hand capable of detecting multiple levels of communities in deep hierarchical networks. Moreover, we propose a strict definition for the term "plateau," which has been always loosely used in previous literature, to help us remove random and irrelevant outputs automatically--without any artificial selection. As a result, our method has "neat" outputs that include only stable and informative plateaus. On synthetic networks with prearranged community structures, our method outperforms most previous methods by recreating the prearranged communities accurately, while on real-world networks, it discovers reasonable community structures that fit the ground truth we already know about the network.