Social Network Analysis, Community Detection Algorithms, and Neighbourhood Identification in Pompeii

The definition and identification of urban neighbourhoods in archaeological contexts remain complex and problematic, both theoretically and empirically. As constructs with both social and spatial characteristics, their detection through material culture alone remains elusive, especially within large settlements that are incompletely excavated or preserved. Thanks to its focus upon relational ties, network analysis offers a profitable path towards untangling the complexities of urban neighbourhoods, especially with respect to their often imprecise, fuzzy boundaries. Various community detection algorithms offer mathematical solutions for partitioning large graphs into communities, but these should not be applied without careful interpretation. Two of the most widely utilized community detection algorithms based on modularity optimization, Louvain and Leiden, contain a customizable resolution parameter that is often overlooked by practitioners. This controls the density of the partitioned communities, and therefore the number identified, but it is difficult to determine the optimal value for any given network. In addition, the results of community detection algorithms vary stochastically. Reliance upon a single iteration may mask potentially significant differences between runs using even a constant resolution parameter. A recently developed algorithm, the Convex Hull of Admissible Modularity Partitions (CHAMP), is designed to overcome these complications and also generates potentially useful multiscalar network community structures. Its applicability to neighbourhood archaeology is demonstrated within three networks of Pompeian housing units based on shared public fountains.

The case study examines Pompeii’s public fountains as hubs of social interaction. Given their daily frequentation by nearby inhabitants, fountains represent plausible proxies for the centres of definable neighbourhoods. It expands upon a spatial network model that connected all 2000+ external doors in the city to the 40 public fountains that were likely functional in 79 CE. Three undirected, one-mode networks were constructed in which units are linked to each other by a common fountain and weighted by the number of fountains they share. The first network connects units by the closest fountain to any external door. Since many properties had side doors within reach of different water sources, these represent potential interconnections between communities. The second and third networks use incrementally larger time to fountain thresholds (30-second and one-minute walks to any fountain, respectively) to map potential choices, also expanding social integration. The results demonstrate that innovative methods for assessing the output of community detection algorithms offer new modes of analysis that are applicable not just to neighbourhood archaeology, but any archaeological network analysis that uses graph partitioning.