New generative model for temporal networks uses mutual information to control community evolution

A new paper from Peijie Zhong, Raúl Mondragón, and Richard Clegg presents a generative model for temporal networks that offers fine-grained control over community structure evolution. Unlike static network models, this approach represents community structure as a sequence of partitions over time. It employs a genetic search algorithm guided by a similarity measure based on mutual information to regulate how communities change between snapshots. This allows explicit control of community evolution including splits and merges, while also handling node additions and removals—a key challenge in real-world dynamic networks like social media or communication graphs.

Temporal edges are generated using intra- and inter-community probabilities derived from either empirical data or theoretical bounds, ensuring connectivity across time steps. The authors validated the model on real-world datasets, demonstrating its ability to replicate the evolution of genuine dynamic communities. Importantly, they then used the model as a benchmark to study the impact of node join/leave rates on the performance of dynamic community detection algorithms. The code and data are linked via arXiv, making it accessible for further research. This work provides a much-needed tool for testing and comparing algorithms in a controlled yet realistic setting.