Bagging Evolutionary ROC-based Hypotheses Application to Terminology Extraction
Jérôme Azé, Mathieu Roche, Michèle Sebag

Abstract

The claim of the paper is that Evolutionary Learning is a source of diverse hypotheses “for free”, and this specificity can be used to combine in an ensemble the hypotheses learned in independent runs. The aim of our algorithm named BROGER (Bagging-ROC GEnetic LEarneR) consists of optimizing the Area Under the ROC Curve using Evolutionary Learning. This paper first presents the theoretical framework of BROGER and then its application to a Term Extraction task in Text Mining