Community Detection in Complex Networks

Abstract

This thesis advances the study of complex network analysis by introducing novel methodologies based on Closed Trail (CT) distance and k-CT components. These innovations extend traditional network metrics by incorporating cyclical dependencies among nodes, offering a more nuanced understanding of network structures and enhancing the detection of overlapping communities. The research begins by developing higher-order clustering and closure coefficients derived from k-CT components. These coefficients provide a refined perspective on local and global network structures. Empirical evaluations demonstrate their ability to distinguish between different network types and yield deeper insights into the underlying network organization. Furthermore, this thesis introduces the Graph Hierarchical Agglomerative Clustering (GHAC) method, which uses cliques or k-CT components as building blocks for detecting overlapping communities. The GHAC method employs CT-based dissimilarity measures, effectively identifying overlapping communities in complex networks. Further refinement of the GHAC method includes its extension to weighted networks, broadening its applicability to various real-world scenarios. Empirical validation of the proposed methodologies shows that the GHAC method performs comparably or superiorly to established algorithms regarding community detection quality, particularly in scenarios involving overlapping communities. The method's application to real-world datasets, such as OECD trade networks, underscores its practical utility and effectiveness in uncovering hierarchical community structures.