Methods for the analysis of complex group formation mechanisms in attributed networks

Community detection has become a fundamental tool for the analysis of complex networks, yet many real-world systems challenge the assumptions underlying traditional approaches. Strong degree heterogeneity, the presence of influential actors, and the availability of node attributes often limit or question the effectiveness of purely topology-based methods, which tend to overlook the mechanisms driving group formation. Addressing these limitations requires frameworks that explicitly account for distinct structural roles and aggregation driven by attribute similarity. Community detection in heterogeneous and attributed networks can be approached from different perspectives. Methods focused on standard definitions of communities emphasise connectivity patterns but struggle in the presence of hubs and high inter-community connectivity, while attribute-based approaches capture similarity but may underweight structural organisation. This thesis adopts an integrated perspective and treats degree heterogeneity and leadership as informative features rather than distortions. By doing so, the proposed framework advances an interpretable view of network organisation that extends beyond individual nodes to explicitly capture community-level structure and hierarchy, highlighting how leadership and connectivity patterns shape the formation, interaction, and positioning of communities within the broader network. This thesis draws on a diverse set of large-scale relational datasets spanning labour flow, housing, and scientific collaboration networks. Empirical analyses are conducted using regional administrative data on worker mobility, spatially explicit housing market data, European research collaboration data derived from the Horizon 2020 and Horizon Europe programmes, and bibliometric co-authorship data on Italian academic scholars constructed from the Italian Ministry of University and Research (MUR) registry and Scopus publication records. Three main contributions are presented. First, empirical analyses demonstrate how structurally prominent actors shape group formation across different applied contexts. Second, a density-based community detection framework is extended to integrate structural information and node attributes, allowing homophily to stabilise community boundaries and redefine leadership in terms of attribute-based community representativeness. Finally, a community-level approach to core-periphery detection is introduced, highlighting the role of peripheral community leaders in maintaining connectivity within the overall system.

Community detection has become a fundamental tool for the analysis of complex networks, yet many real-world systems challenge the assumptions underlying traditional approaches. Strong degree heterogeneity, the presence of influential actors, and the availability of node attributes often limit or question the effectiveness of purely topology-based methods, which tend to overlook the mechanisms driving group formation. Addressing these limitations requires frameworks that explicitly account for distinct structural roles and aggregation driven by attribute similarity. Community detection in heterogeneous and attributed networks can be approached from different perspectives. Methods focused on standard definitions of communities emphasise connectivity patterns but struggle in the presence of hubs and high inter-community connectivity, while attribute-based approaches capture similarity but may underweight structural organisation. This thesis adopts an integrated perspective and treats degree heterogeneity and leadership as informative features rather than distortions. By doing so, the proposed framework advances an interpretable view of network organisation that extends beyond individual nodes to explicitly capture community-level structure and hierarchy, highlighting how leadership and connectivity patterns shape the formation, interaction, and positioning of communities within the broader network. This thesis draws on a diverse set of large-scale relational datasets spanning labour flow, housing, and scientific collaboration networks. Empirical analyses are conducted using regional administrative data on worker mobility, spatially explicit housing market data, European research collaboration data derived from the Horizon 2020 and Horizon Europe programmes, and bibliometric co-authorship data on Italian academic scholars constructed from the Italian Ministry of University and Research (MUR) registry and Scopus publication records. Three main contributions are presented. First, empirical analyses demonstrate how structurally prominent actors shape group formation across different applied contexts. Second, a density-based community detection framework is extended to integrate structural information and node attributes, allowing homophily to stabilise community boundaries and redefine leadership in terms of attribute-based community representativeness. Finally, a community-level approach to core-periphery detection is introduced, highlighting the role of peripheral community leaders in maintaining connectivity within the overall system.