Ontology-Based Classification – Application of Machine Learning Concepts Without Learning

Abstract

The application of machine learning algorithms on real world problems rarely encounters ideal conditions. Often either the available data are imperfect or insufficient, or the learning situation requires a rather complicated combination of different approaches. In this article I describe an application, which – in an ideal world – would be solvable by a conventional supervised classification algorithm. Unfortunately, the available data are neither reliably classified nor could a manual correct reclassification be derived under restricted available resources. Since we had a large domain thesaurus available, we were able to develop a new approach, skipping the learning step and deriving a classification model directly from this thesaurus. The evaluation showed that for the intended use the quality of the classification model is more than sufficient.