Central node identification via weighted kernel density estimation

The detection of central nodes in a network is a fundamental task in network science and graph data analysis. During the past decades, numerous centrality measures have been presented to characterize what is a central node. However, few studies address this issue from a statistical inference perspective. In this paper, we formulate the central node identification issue as a weighted kernel density estimation problem on graphs. Such a formulation provides a generic framework for recognizing central nodes. On one hand, some existing centrality evaluation metrics can be unified under this framework through the manipulation of kernel functions. On the other hand, more effective methods for node centrality assessment can be developed based on proper weighting coefficient specification. Experimental results on 20 simulated networks and 53 real networks show that our method outperforms both six prior state-of-the-art centrality measures and two recently proposed centrality evaluation methods. To the best of our knowledge, this is the first piece of work that addresses the central node identification issue via weighted kernel density estimation.