Evaluation of Word Vectors by Synonym Identification - Skip-gram Word Vectors as an Example -

Abstract

Automatic synonym acquisition is an important problem in the field of document retrieval and data mining using natural language data. In this paper, we conducted two experiments to identify the properties of word vectors acquired by the Skip-gram model, related to the synonym identification. In the first experiment, we confirmed that the cosine similarity of a synonym pair is significantly higher than that of a non-synonym pair. In the second experiment, we show that only a limited number of components of word vectors are needed for discriminating synonym pairs from non-synonym pairs.