Skeleton coupling: a novel interlayer mapping of community evolution in temporal networks

TL;DR

This paper introduces skeleton coupling, a new method for interlayer linking in temporal networks, which improves dynamic community detection especially in networks with transient and singleton communities, such as neuronal data.

Contribution

The paper proposes skeleton coupling, a novel interlayer link definition that enhances community tracking in temporal networks with transient structures.

Findings

Skeleton coupling improves DCD performance in synthetic neuronal networks.

Interlayer link topology significantly affects community detection accuracy.

Integrating skeleton coupling yields more interpretable community evolution results.

Abstract

Dynamic community detection (DCD) in temporal networks is a complicated task that involves the selection of a method and its associated hyperparameters. How to choose the most appropriate method generally depends on the type of network being analyzed and the specific properties of the data that define the network. In functional temporal networks derived from neuronal spike train data, communities are expected to be transient, and it is common for the network to contain multiple singleton communities. Here, we compare the performance of different DCD methods on functional temporal networks built from synthetic neuronal time series data with known community structure. We find that, for these networks, DCD methods that utilize interlayer links to perform community carryover between layers outperform other methods. However, we also observe that DCD performance is highly dependent on the topology of interlayer links, especially in the presence of singleton and transient communities. We therefore define a novel way of defining interlayer links in temporal networks called skeleton coupling that is specifically designed to enhance the linkage of communities in the network throughout time based on the topological properties of the community history. We show that integrating skeleton coupling with current DCD methods improves the method's performance in synthetic data with planted singleton and transient communities. The use of skeleton coupling to perform DCD will therefore allow for more accurate and interpretable results of community evolution in real-world neuronal data or in other systems with transient structure and singleton communities.