Improving word and sense embedding with hierarchical semantic relations
Yow-Ting Shiue, Wei-Yun Ma · 2017
Distributed representations are shown useful in many NLP tasks. Besides the context, lexical resources also provide valuable information about lexical units. This paper proposes simple methods to improve word and sense vectors by training on multi-level hierarchical semantic relations. Experiments on both intrinsic and extrinsic tasks show that our approach leads to consistent improvement competitive with state-of-the-arts. Moreover, the enhanced vectors are directly applicable to existing applications.