Meronymy Relation Extraction Based on 3-Motif in Wikidata

Abstract

Meronymy relation extraction is playing an important role in knowledge mining and it is a huge challenge to extract meronymy relation from a large amount of complicated unlabeled data. In Wikidata, it is found that meronymy relations appear in some specific network topologies of the hyperlinks. Using this characteristic of Wikidata, we proposed a new approach to extract meronymy relations: based on all 13 different three-node motifs, we build a 13-dimensional feature vector for each hyperlink to be classified with different classification algorithms. We used part of labeled data in Wikidata to train models and used the left data to judge the quality of this classification model. With some of the classification algorithms, our new approach can extract meronymy relations in higher accuracy and works well in diverse fields.