On kernel methods for covariates that are rankings

Kernel methods provide an attractive framework for aggregating and learning from ranking data, and so understanding the fundamental properties of kernels over permutations is a question of broad interest. We provide a detailed analysis of the Fourier spectra of the standard Kendall and Mallows kernels, and a new class of polynomial-type kernels. We prove that the Kendall kernel has exactly two irreducible representations at which the Fourier transform is non-zero, and moreover, the associated matrices are rank one. This implies that the Kendall kernel is nearly degenerate, with limited expressive and discriminative power. In sharp contrast, we prove that the Fourier transform of the Mallows kernel is a strictly positive definite matrix at all irreducible representations. This property guarantees that the Mallows kernel is both characteristic and universal. We introduce a family of normalized polynomial kernels of degree p that interpolates between the Kendall (degree one) and Mallows (infinite degree) kernels, and show that for d-dimensional permutations, the p-degree kernel is characteristic when p ≥ d − 1, unlike the Euclidean case in which no finite-degree polynomial kernel is characteristic.