TEMP: Taxonomy Expansion with Dynamic Margin Loss through Taxonomy-Paths

• DOI: 10.18653/v1/2021.emnlp-main.313
• Publication Date: 2021-12-17T03:56:42Z
• Authors: Zichen Liu; Hongyuan Xu; Yanlong Wen; Ning Jiang; HaiYing Wu; Xiaojie Yuan
• Keywords: Margin (machine learning); Mean reciprocal rank; Rank (graph theory); Representation

As an essential form of knowledge representation, taxonomies are widely used in various downstream natural language processing tasks. However, with the continuously rising new concepts, many existing unable to maintain coverage by manual expansion. In this paper, we propose TEMP, a self-supervised taxonomy expansion method, which predicts position concepts ranking generated taxonomy-paths. For first time, TEMP employs pre-trained contextual encoders construction and hypernym detection problems. Experiments prove that embeddings able capture hypernym-hyponym relations. To learn more detailed differences between taxonomy-paths, train model dynamic margin loss novel function. Extensive evaluations exhibit outperforms prior state-of-the-art approaches 14.3% accuracy 15.8% mean reciprocal rank on three public benchmarks.