Detecting groups of nodes in large real-world networks using label propagation

Lovro Šubelj (2013) Detecting groups of nodes in large real-world networks using label propagation. PhD thesis.

Preview

PDF Download (19Mb)

Abstract

The World Wide Web, wiring of a neural system, “Facebook” and a plumbing are all examples of complex networks composed of a large number of interconnected components denoted nodes. Many such real-world networks reveal characteristic patterns of connectedness that are far from regular or random. Analysis of the structure of large real-world networks is thus an active research field in physics, mathematics, computer science and informatics, whereas network analysis also represents the foundations of many successful companies (e.g., “Google”). Present doctoral thesis focuses on the development of approaches for detection of characteristic groups of nodes in large complex networks. We present different group detection methods and techniques based on label propagation that improve its robustness and accuracy, and extend the approach to more general groups of nodes. Presented advances represent a complete solution for group detection. In contrast to most other approaches, ours do not require any prior knowledge about the structure of the network. Performance on synthetic and real-world networks with known structure is at least comparable to current state-of-the-art approaches, while the asymptotic complexity is near ideal. Furthermore, presented approaches allow simple comprehension and implementation, and also a straight-forward incorporation of arbitrary knowledge about the underlying domain. Note that the above-mentioned properties are met by only a few other approaches in the literature. The doctoral thesis also focuses on the analysis of groups of nodes in software networks. Complex software is probably one of the most sophisticated human-made systems, however, only little is known about the actual structure of high-quality software projects. Analyses show that software networks contain significant groups of nodes that are relatively well depicted in the network structure, whereas the group hierarchy also roughly coincides with the organization of the corresponding program libraries. Based on the latter, we present different practical applications of group detection in software engineering. Doctoral thesis is based on four published papers that together constitute the core of the thesis, and are included in the same form as published.

Actions (login required)