Learning to learn biological relations from a small training set

Abstract

In this paper we present different ways to improve a basic machine learning approach to identify relations between biological named entities as annotated in the Genia corpus. The main difficulty with learning from the Genia event-annotated corpus is the small amount of examples that are available for each relation type. We compare different ways to address the data sparseness problem: using the corpus as the initial seed of a bootstrapping procedure, generalizing classes of relations via the Genia ontology and generalizing classes via clustering. © Springer-Verlag Berlin Heidelberg 2009.