Comparison of algorithms for network community detection based on label propagationVAREŠANOVIĆ, DAMIR (Avtor) Šubelj, Lovro (Mentor) Complex networkCommunity detectionLabel propagationConsensus clusteringProgramming libraryCommunity structure is an important property of complex networks, since it reveals the organization of the network and relationships between its members. Therefore, the analysis of community structure and development of effective procedures for its detection has been one of the main focuses of network theory. Numerous methods have been proposed for detecting community structure in networks \cite{article7}. This thesis presents a heuristic community detection algorithm based on label propagation. Due to its simplicity and low time complexity, label propagation algorithm should be the first option to provide a better understanding of the network community structure before examining other more complex alternatives. We give a brief introduction to graphs and networks, different clustering metrics and related work in the field of network community detection. Next, we present the basic approach of label propagation algorithm, discuss advantages and disadvantages, and review extensions of the method, focusing mainly on consensus clustering and fast consensus clustering. The aforementioned algorithms are implemented in a Python programming library, which is available at: \url{https://github.com/damir1407/label-propagation}. Furthermore, we evaluate these three network clustering methods on different synthetic and real-world networks, and present the results. The thesis is concluded with a summary of the presented methods and directions for future work.20212021-02-15 08:24:04Diplomsko delo/naloga124752sl