IbLT: An effective granular computing framework for hierarchical community detection

摘要

Mapping the vertices of network onto a tree helps to reveal the hierarchical community structures. The leading tree is a granular computing (GrC) model for efficient hierarchical clustering and it requires two elements: the distance between granules, and the density calculated in Euclidean space. For the non-Euclidean network data, the vertices need to be embedded in the Euclidean space before density calculation. This results in the marginalization of community centers. This paper proposes a new hierarchical community detection framework, called Importance-based Leading Tree (IbLT). Different from the density-based leading tree, IbLT calculates the structural similarity between vertices and the importance of the vertices respectively. It generates leading trees that match the structural features of the vertices, and thus, IbLT obtains more accurate results for the detection of hierarchical community structures. Experiments are conducted to evaluate the performance of the proposed novel IbLT-based method. On social network community detection task, the quantitative results show that this method achieves competitive accuracy.