DynBenchmark: Customizable Ground Truths to Benchmark Community Detection and Tracking in Temporal Networks

TL;DR

DynBenchmark introduces a highly configurable benchmark for dynamic community detection in temporal networks by generating evolving communities and corresponding SBM-based underlying graphs with controllable lifespans, births, and member flows. The framework evaluates algorithms along three axes—partition quality, transition tracking, and evolutionary event detection—using ground-truth-contingency structures and ICEM-based event classification, enabling nuanced performance insights. Experiments with Louvain, Infomap, and Walktrap reveal distinct strengths and failures across topology and dynamics (e.g., Infomap’s event-detection strengths vs. partition tracking weaknesses). The package provides open-source generation, validation metrics, and visualization tools, offering a practical platform for standardized, multifaceted evaluation and method selection in real-world temporal networks.

Abstract

Graph models help understand network dynamics and evolution. Creating graphs with controlled topology and embedded partitions is a common strategy for evaluating community detection algorithms. However, existing benchmarks often overlook the need to track the evolution of communities in real-world networks. To address this, a new community-centered model is proposed to generate customizable evolving community structures where communities can grow, shrink, merge, split, appear or disappear. This benchmark also generates the underlying temporal network, where nodes can appear, disappear, or move between communities. The benchmark has been used to test three methods, measuring their performance in tracking nodes' cluster membership and detecting community evolution. Python libraries, drawing utilities, and validation metrics are provided to compare ground truth with algorithm results for detecting dynamic communities.

Figures (6)

• Figure 1: Communities' lifespan and interactions. Colors differentiate communities, columns reflect time (10 snapshots). The thickness or thinness of the gray flows is indicative of the number of members migrating between communities.
• Figure 2: Impact of core node ratio on members' trajectories: number of communities visited by each node and their lifetime
• Figure 3: Properties of the underlying networks
• Figure 4: (a) Evolution of Louvain's performance (NMI) over time with fixed internal connectivity ($p_{in}=0.5$) and different values of $p_{out}$. (b-d) Sensitivity analysis showing the impact of both $p_{in}$ and $p_{out}$ on the performance of Louvain, Infomap, and Walktrap respectively.
• Figure 5: Temporal tracking performance