Feature-based approach combined with hierarchical classifying strategy to relation extraction

Abstract

This paper proposes a novel feature-based method for relation extraction task. Diverse lexical and syntactic features are defined to describe the context of the pair of entities. Dependency features are selected to capture the structure and dependency information of sentence. Hierarchical classifying strategy is used to reduce the weakness of the traditional approach, which treats training examples in different classes equally and independently, At the same time, correction mechanism is used to improve the performance of the system.