On the Effectiveness of Using Link Weights and Link Direction for Community Detection in Multilayer Networks

Multilayer networks are useful representations of real-world complex networks, and community detection in multilayer networks has been an area of active research. There are several options for the graph representation of a multilayer network, for example, directed or undirected and weighted or unweighted. Although these options may affect the results of community detection in a multilayer network, the representations that are effective for community detection have been not yet been clarified. In this paper, we experimentally investigate how the graph representation of a multilayer network affects the results of community detection. Through experiments using multilayer networks of Twitter users, we show that using a directed graph for each layer of a multilayer network contributes to improved accuracy in estimating the communities of Twitter users. We also show that when there is a clear oppositional structure among nodes in a network, manipulating link weights to emphasize the oppositional structure improves the accuracy of community detection.