The use of syntactic structure in relationship extraction

Abstract

This thesis describes a method of relationship extraction that uses surface and syntactic features of text along with entity detection information to perform sentence-level relationship extraction tasks. The tasks are to build and test classifiers for the following relationships: employer-employee, organization-location, family, and person-location. Methods of reducing noise in these features sets are also discussed, and experimental evidence of their effectiveness is presented. Performance of the system is analyzed in terms of precision and recall, and errors of the system are discussed along with proposed solutions. Finally a reformulation of the problem is presented along with a discussion of future work.