Online detection of temporal communities in evolving networks by estrangement confinement

TL;DR

This paper introduces a novel method called estrangement confinement for detecting temporal communities in evolving networks, emphasizing the importance of temporal smoothness and node relationship inertia to improve community detection accuracy.

Contribution

The paper proposes a new measure of partition distance called estrangement and integrates it into community detection, enabling more reliable identification of temporal communities across diverse datasets.

Findings

Effective detection of temporal communities in real-world datasets

The estrangement confinement method improves community stability over time

Applicable to synthetic and real-world evolving networks

Abstract

Temporal communities result from a consistent partitioning of nodes across multiple snapshots of an evolving complex network that can help uncover how dense clusters in a network emerge, combine, split and decay with time. Current methods for finding communities in a single snapshot are not straightforwardly generalizable to finding temporal communities since the quality functions used for finding static communities have highly degenerate landscapes, and the eventual partition chosen among the many partitions of similar quality is highly sensitive to small changes in the network. To reliably detect temporal communities we need not only to find a good community partition in a given snapshot but also ensure that it bears some similarity to the partition(s) found in immediately preceding snapshots. We present a new measure of partition distance called "estrangement" motivated by the inertia of inter-node relationships which, when incorporated into the measurement of partition quality, facilitates the detection of meaningful temporal communities. Specifically, we propose the estrangement confinement method, which postulates that neighboring nodes in a community prefer to continue to share community affiliation as the network evolves. Constraining estrangement enables us to find meaningful temporal communities at various degrees of temporal smoothness in diverse real-world datasets. Specifically, we study the evolution of voting behavior of senators in the United States Congress, the evolution of proximity in human mobility datasets, and the detection of evolving communities in synthetic networks that are otherwise hard to find. Estrangement confinement thus provides a principled approach to uncovering temporal communities in evolving networks.