Improving Relation Extraction by Using an Ontology Class Hierarchy Feature

Relation extraction is a key step to address the problem of structuring natural language text. This paper proposes a new ontology class hierarchy feature to improve relation extraction when applying a method based on the distant supervision approach. It argues in favour of the expressiveness of the feature, in multi-class perceptrons, by experimentally showing its effectiveness when compared with combinations of (regular) lexical features.