0-Bit Consistent Weighted Sampling

P. Li. KDD 2015


We develop 0-bit consistent weighted sampling (CWS) for efficiently estimating min-max kernel, which is a generalization of the resemblance kernel originally designed for binary data. Because the estimator of 0-bit CWS constitutes a positive definite kernel, this method can be naturally applied to large-scale data mining problems. Basically, if we feed the sampled data from 0-bit CWS to a highly efficient linear classifier (e.g., linear SVM), we effectively (and approximately) train a nonlinear classifier based on the min-max kernel. The accuracy improves as we increase the sample size.

In this paper, we first demonstrate, through an extensive classification study using kernel machines, that the min-max kernel often provides an effective measure of similarity for nonnegative data. This helps justify the use of min-max kernel. However, as the min-max kernel is nonlinear and might be difficult to be used for industrial applications with massive data, we propose to linearize the min-max kernel via 0-bit CWS, a simplification of the original CWS method.

The previous remarkable work on consistent weighted sampling (CWS) produces samples in the form of (i, t) where the i* records the location (and in fact also the weights) information analogous to the samples produced by classical minwise hashing on binary data. Because the t* is theoretically unbounded, it was not immediately clear how to effectively implement CWS for building large-scale linear classifiers. We provide a simple solution by discarding t* (which we refer to as the “0-bit” scheme). Via an extensive empirical study, we show that this 0-bit scheme does not lose essential information. We then apply 0-bit CWS for building linear classifiers to approximate min-max kernel classifiers, as extensively validated on a wide range of public datasets.

We expect this work will generate interests among data mining practitioners who would like to efficiently utilize the nonlinear information of non-binary and nonnegative data.