A framework for tackling myopia in concept learning on the web of data

Abstract

A prominent class of supervised methods for the representations adopted in the context of the Web of Data are designed to solve concept learning problems. Such methods aim at approximating an intensional definition for a target concept from a set of individuals of a target knowledge base. In this scenario, most of the well-known solutions exploit a separate-and-conquer approach: intuitively, the learning algorithm builds an intensional definition by repeatedly specializing a partial solution with the aim of covering the largest number of positive examples as possible. Essentially such a strategy can be regarded as a form of hill-climbing search that can produce sub-optimal solutions. To cope with this problem, we propose a novel framework for the concept learning problem called DL-Focl. Three versions of this algorithmic solution, built upon DL-Foil, have been designed to tackle the inherent myopia of the separate-and-conquer strategies. Their implementation has been empirically tested against methods available in the DL-Learner suite showing interesting results.