This paper develops an approximation to the (effective) -resistance and applies it to multi-class clustering. Spectral methods based on the graph Laplacian and its generalization to the graph -Laplacian have been a backbone of non-euclidean clustering techniques. The advantage of the -Laplacian is that the parameter induces a controllable bias on cluster structure. The drawback of -Laplacian eigenvector based methods is that the third and higher eigenvectors are difficult to compute. Thus, instead, we are motivated to use the -resistance induced by the -Laplacian for clustering. For -resistance, small biases towards clusters with high internal connectivity while large biases towards clusters of small "extent," that is a preference for smaller shortest-path distances between vertices in the cluster. However, the -resistance is expensive to compute. We overcome this by developing an approximation to the -resistance. We prove upper and lower bounds on this approximation and observe that it is exact when the graph is a tree. We also provide theoretical justification for the use of -resistance for clustering. Finally, we provide experiments comparing our approximated -resistance clustering to other -Laplacian based methods.
View on arXiv