Home

Welcome to the CommunityFitNet wiki!

This page is a companion for our paper on overfitting and underfitting of community detection methods on real networks, written by Amir Ghasemian, Homa Hosseinmardi, and Aaron Clauset. The page hosts the partitions created by each method in our comparison set of community detection algorithms (Table 1 of the paper).

Our goal is to provide a community detection benchmark used as a standardized reference for comparing community detection methods. The proposed package here contains both the benchmark and the partitions for each included network obtained by the algorithms evaluated in our paper. To compare a new algorithm with those in our evaluation set of algorithms, a researcher can run the new algorithm on the proposed benchmark, and identify which reference algorithm has the most similar behavior, e.g., in the average number of communities found (Fig. 2 of the paper). We belive the availability of this benchmark and the results of running so many state-of-the-art algorithms on it should facilitate further advances in developing community detection algorithms.

Reference:

Amir Ghasemian, Homa Hosseinmardi, and Aaron Clauset,
"Evaluating and Comparing Overfit in Models of Network Community Structure (2017)".