Evaluating community detection algorithms for progressively evolving graphs

Abstract

Many algorithms have been proposed in the last 10 years for the discovery of dynamic communities. However, these methods are seldom compared between themselves. In this article, we propose a generator of dynamic graphs with planted evolving community structure, as a benchmark to compare and evaluate such algorithms. Unlike previously proposed benchmarks, it is able to specify any desired evolving community structure through a descriptive language, and then to generate the corresponding progressively evolving network. We empirically evaluate six existing algorithms for dynamic community detection in terms of instantaneous and longitudinal similarity with the planted ground truth, smoothness of dynamic partitions and scalability. We notably observe different types of weaknesses depending on their approach to ensure smoothness, namely Glitches, Oversimplification and Identity loss. Although no method arises as a clear winner, we observe clear differences between methods, and we identified the fastest, those yielding the most smoothed or the most accurate solutions at each step.

Users’ polarisation in dynamic discussion networks: The case of refugee crisis in Sweden

“…The algorithms differ in terms of both community detection in each of the network snapshots (i.e., in each of 96 static networks) and the community evolution tracking procedure. In particular, we have evaluated smoothed Louvain [ 101 ] and Clauset-Newman-Moore greedy modularity maximisation [ 102 ] community detection methods, along with community matching in consecutive snapshots [based on the rules described in 97 ], label smoothing [ 99 ] and smoothed Louvain, where community detection takes place based on the partition on a previous time-step [ 100 ]. We evaluated dynamic partition based on the modularity at each step, consecutive similarity and global smoothness scores, such as the average value of partition smoothness, node smoothness and label smoothness [ 97 ].…”

Comparison between Louvain and Leiden Algorithm for Network Structure: A Review

“…However, for the agglomerative methods (refer right wing), is the other way round where from down to up. It means that from the complex network that content all nodes, they will iteratively merge if similarity becomes a high so-called merged process [8]. This paper shows the Louvain and Leiden algorithm are categories in agglomerative method.…”

Dynamic configuration of large-scale cortical networks during an inhibitory task accounts for heterogeneity in attention-deficit/hyperactivity disorder traits

“…Dynamic modules were detected using the method described by Aynaud & Guillaume (57), which applies the Louvain algorithm to each static network at time t, except, rather than the algorithm first assigning each node to their own module, nodes are first assigned to the module with which they belonged at t -1. This avoids issues that arise from community-detection methods which identify modules at each timepoint independently (58). Due to the small variance that arises from Louvain's algorithm, flexibility was averaged across 100 runs (59).…”