Word Vector Computation Based on Implicit Expression

Abstract

Word vector and topic model can help retrieve information semantically to some extent. However, there are still many problems. (1) Antonyms share high similarity when clustering with word vectors. (2) Number of all kinds of name entities, such as person name, location name, and organization name is infinite while the number of one specific name entity in corpus is limited. As the result, the vectors for these name entities are not fully trained. In order to overcome above problems, this paper proposes a word vector computation model based on implicit expression. Words with the same meaning are implicitly expression based on dictionary and part of speech. With the implicit expression, the sparsity of corpus is reduced, and word vectors are trained deeper.