Link Fraction Mixed Membership Reveals Community Diversity in Aggregated Social Networks
Netherlands study shows urban hubs as melting pots of remote communities.
Community detection has long struggled with aggregated or coarse-grained social networks, where disjoint partitions miss the diversity of memberships within aggregated nodes. Existing mixed membership methods can capture this diversity but are often highly sensitive to aggregation resolution, leading to inconsistent results. In a new preprint on arXiv (2602.03266), researchers Gamal Adel, Eszter Bokányi, Eelke M. Heemskerk, and Frank W. Takes introduce Link Fraction Mixed Membership (LFMM), a method that computes mixed memberships in aggregated networks while remaining consistent under aggregation—meaning it conserves community membership sums at different scales.
Applying LFMM to a population-scale social network of the Netherlands, the team analyzed data aggregated at different geographical resolutions over the past decade. The method revealed clear variation in community membership across regions and showed how large urban hubs act as melting pots, bringing together members from spatially remote communities. This work, supported by 21 pages and 6 figures, offers a robust way to study the mesoscopic structure of large-scale networks without losing community diversity, with implications for urban planning, migration studies, and network science.
- LFMM is the first mixed membership method that is consistent under aggregation, preserving community membership sums across different scales.
- Applied to a population-scale social network of the Netherlands, covering multiple geographical resolutions and a decade of evolution.
- Identified large urban hubs as melting pots that integrate diverse community members from spatially remote regions.
Why It Matters
Enables robust community analysis of large aggregated social networks, revealing real diversity patterns like urban melting pot effects.