关键词: 文本蕴含, 知识向量, 深度学习, 词对齐, 注意力机制

Abstract:

Text entailment technology has been widely used in natural language processing, but there are some problems such as the poor reasoning ability of word pairs （for example, there is the antonym pairs in sentence pairs, but can’t judge the antonym relationship, etc.）. This paper focuses on the acquisition of knowledge vector from words, including acquire the word pairs relation vector by integration of multi feature and supervised method, acquire word pairs relation expression using TransR tools, and acquire antonym vector expression. Knowledge vector is introduced into the part of word alignment and attention mechanism in text entailment recognition model. The experimental results show that the proposed method is better than the classical model.

Key words: text entailment, knowledge representation, deep learning, word alignment, attention mechanism