We introduce the Convex Hull of Admissible Modularity Partitions (CHAMP)algorithm to prune and prioritize different network community structuresidentified across multiple runs of possibly various computational heuristics.Given a set of partitions, CHAMP identifies the domain of modularityoptimization for each partition ---i.e., the parameter-space domain where ithas the largest modularity relative to the input set---discarding partitionswith empty domains to obtain the subset of partitions that are "admissible"candidate community structures that remain potentially optimal over indicatedparameter domains. Importantly, CHAMP can be used for multi-dimensionalparameter spaces, such as those for multilayer networks where one includes aresolution parameter and interlayer coupling. Using the results from CHAMP, auser can more appropriately select robust community structures by observing thesizes of domains of optimization and the pairwise comparisons betweenpartitions in the admissible subset. We demonstrate the utility of CHAMP withseveral example networks. In these examples, CHAMP focuses attention ontopruned subsets of admissible partitions that are 20-to-1785 times smaller thanthe sets of unique partitions obtained by community detection heuristics thatwere input into CHAMP.
展开▼