Automatic Hypernym-Hyponym Relation Extraction With WordNet Projection

Congcong Zhang, Gaofei Xie, Ning Liu, Xiaojie Hu, Yatian Shen, Xiajiong Shen

Abstract: Semantic taxonomies such as WordNet are expensive to build manually. Based on the above ideas and problems, in this paper, we propose a new model for automatically extracting hypernym-hyponym relations from raw text. Our model is based on two scoring functions, one neural networks model for sentence including word pair and the other model used projection network, which learns semantic distance to noun pairs in semantic hierarchy of WordNet. At last, sum of two score is used to classify hypernym-hyponym relation from raw text. On the experimental tasks about hypernym-hyponym relation extraction, our methods get about the result of 0.82 F1-score.

External IDs:dblp:conf/icsai/ZhangXLHSS21a