Incorporating lexical semantic similarity to tree kernel-based chinese relation extraction

Lexical semantic information plays an important role in semantic relation extraction between named entities. This paper incorporates two kinds of lexical semantic similarity measures, thesaurus-based and corpus-based, into convolution tree kernels and systematically investigates their effects on Chinese relation extraction. The experiments on the ACE2005 Chinese corpus shows that the incorporation of lexical semantic similarity into tree kernel-based Chinese relation extraction can significantly improve the extraction performance when entity types are unknown, while in the case of known entity types, these lexical similarity measures also enhance the extraction performance for some person-related relationships. This demonstrates the usefulness of lexical semantic similarity in Chinese relation extraction.