Combining Syntactic Structured and Flat Features for Relation Extraction Using Co-training

Abstract

A parse tree contains rich syntactic structured information, and the structured features have been proved effective in relation extraction. In this paper, we proposed another way to efficiently utilize structured features but in a weakly learning way. Co-training algorithm was chosen by us, the structured features were set to be one view of it, and the flat features were set to be the other. Through using co-training algorithm, we can combine both flat and structured information for relation extraction.