Evaluating Word Embeddings based on Hypernymy Relations

摘要

Word vector representation is the basis of natural language processing tasks. There are many excellent articles that propose different methods of word vector representation. However, there is still no unified and accepted method for evaluating word vector representation. We present an evaluating embedding method-HEWE (hypernymy evaluates word embeddings) based on a combined hypernymy relation detection that integrates both path-based and distributional features. And prove the method by the experiments, comparing with the sentiment classification result.