Enhancing Relation Extraction by Eliciting Selectional Constraint Features from Wikipedia

Gang Wang, Huajie Zhang, Haofen Wang, Yong Yu

Abstract: Selectional Con straints are usually checked for detecting semantic relations. Previous work usually defined the constraints manually based on hand crafted concept taxonomy, which is time-consuming and impractical for large scale relation extraction. Further, the determination of entity type (e.g. NER) based on the taxonomy cannot achieve sufficiently high accuracy. In this paper, we propose a novel approach to extracting relation instances using the features elicited from Wikipedia, a free online encyclopedia. The features are represented as selectional constraints and further employed to enhance the extrac tion of relations. We conduct case stud ies on the validation of the ex tracted instances for two common relations hasAr tist(album, artist) andhasDirector(film, director). Substantially high extraction precision (around 0.95) and validation accuracy (near 0.90) are obtained.