Unsupervised machine learning for identification of audience subpopulations and dimensionality and/or sparseness reduction techniques to facilitate identification of audience subpopulations

Abstract

Some embodiments described herein relate to a computer-implemented method that includes accessing behavioral data, such as web visitation data, of multiple users. A sparse behavioral vector can be defined for each user based on the behavioral data. Each element of each sparse behavioral vector can represent a different potential detectable behavior such that each sparse behavioral vector encodes the behavioral data for that user. Multiple supervised learning models to each sparse behavioral vector to densify the vectors, defining multiple dense behavioral vectors. An unsupervised machine learning technique can be applied to the dense behavioral vectors to cluster, or define subpopulations, based on similarities between the dense behavioral vectors. Delivery of targeted content to a user can be facilitated based on a dense behavioral vector associated with that user being associated with one or more of the clusters or subpopulations.