Joint Learning of Hierarchical Word Embeddings from a Corpus and a Taxonomy
Mohammed Alsuhaibani, Takanori Maehara, Danushka Bollegala · Automated Knowledge Base Construction · 2019
Identifying the hypernym relations that hold between words is a fundamental task in NLP. Word embedding methods have recently shown some capability to encode hypernymy. However, such methods tend not to explicitly encode the hypernym hierarchy that exists between words. In this paper, we propose a method to learn a hierarchical word embedding in a specii¬c order to capture the hypernymy. To learn the word embeddings, the proposed method considers not only the hypernym relations that exists between words on a taxonomy, but also their contextual information in a large text corpus. The experimental results on a supervised hypernymy detection and a newly-proposed hierarchical path completion tasks show the ability of the proposed method to encode the hierarchy. Moreover, the proposed method outperforms previously proposed methods for learning word and hypernym-specii¬c word embeddings on multiple benchmarks.