Attribute analysis in biomedical text classification

Abstract

Text Classification tasks are becoming increasingly popular in the field of Information Access. Being approached as Machine Learning problems, the definition of suitable attributes for each task is approached in an ad-hoc way. We believe that a more principled framework is required, and we present initial insights on attribute engineering for Text Classification, along with a software library that allows experiment definition and fast prototyping of classification systems. The library is currently being used and evaluated in Information Access projects in the biomedical domain.