Application of Output Embedding on Word2Vec
Shuto Uchida, Tomohiro Yoshikawa, Takeshi Furuhashi · 2018
The word vector of distributed representation that embeds the semantic relationship of words into a vector using Word2Vec has attracted attention in recent years. Furthermore, this word vector has become widely to be used in the field of Natural Language Processing such as parsing and document classification, and its effectiveness has been reported. Generally, it is the input embedding on Word2Vec that is used as the elements of a word vector, and the output embedding generated at the same time is not used. On the other hand, the authors focus on the usefulness of the paired output embedding. In this paper, we propose word vectors using the input and output embeddings together. In addition, we experimentally investigate the performance of the proposed word vectors and carry out document classification experiments using the proposed word vectors. The result shows that the classification performance was improved by the proposed word vectors.