Similarity Metric Based on Resistance Distance and its Applications to Data Clustering

We propose a new clustering method that uses a similarlity metric deived from electrical resistance networks. The proposed metric allows us to quantify the mutual relevancy between data objects or nodes. We show how to derive the metric from the data collection and how to apply it to various application contexts. Our theoretical analyses and experiments show the excellent potential of the method to identifying clusters of networks and to improving data clustering performance on a number of data sets.