Dynamic Community Detection

There is no universal default. Run the algorithm across a grid of omega values and inspect how community count and membership stability respond. Low omega yields rapidly shifting communities; high omega enforces near-constant membership. Choose a value in the range where your substantive findings are stable — this is a form of sensitivity analysis and should be reported.

The terms are often used interchangeably in the literature. Some authors reserve 'temporal' for methods that model continuous-time processes or use timestamps explicitly, while 'dynamic' can also cover networks where the node set itself changes. In practice, both labels refer to detecting communities in networks that change over time.

Yes, but it requires careful treatment. Nodes absent in a given snapshot must be assigned a dummy state or excluded from that layer's modularity calculation. Some implementations handle this natively; others require pre-processing to ensure a consistent node set across snapshots, which may artificially inflate or suppress community sizes.

Louvain is the most common heuristic because it scales well and often finds high-quality partitions, but it is stochastic and can return different results across runs. For reproducibility, fix a random seed, run multiple times, and check that key findings are stable. The Leiden algorithm is a more recent alternative that corrects some of Louvain's internal-connectivity weaknesses.

At a minimum, three snapshots are needed to observe a meaningful temporal trajectory (before, during, after). In practice, studies with fewer than five to ten snapshots often struggle to distinguish genuine community evolution from noise. The ideal number depends on the rate of change in your network; faster-changing systems require finer temporal resolution and thus more snapshots.