dynamicmultiplex - Tracking Communities in Changing Temporal Networks

Tools for temporal networks that change over time, where nodes enter and leave, groups form, split, merge and dissolve, and periods differ in size. The temporal fitters track nodes and communities across periods. A community keeps one identifier while it persists, and a newly formed group receives a new one. Two-stage fitters run Louvain or Leiden community detection in each period and link communities in linked periods by Jaccard similarity, the overlap coefficient or node-strength weighted variants. The 'layer_links' argument sets which periods are directly linked, adjacent periods only by default. By default all periods are fitted jointly, so later periods can change earlier labels, and tracking = 'filtered' prevents this. Multislice modularity (Mucha et al. 2010) <doi:10.1126/science.1184819> and Hungarian matching are included as comparison methods, and a bootstrap stability score helps choose between the two-stage and multislice approaches.