Research on Bi-LSTM machine reading comprehension algorithm based on attention mechanism

Abstract

Abstract Machine reading comprehension is a very important research field in Natural Language Processing (NLP), but it is not sufficient for military research, and traditional machine reading comprehension algorithms do not fully explore every word in different contexts. In response to the above problems, this paper proposes a machine reading comprehension model that incorporates a multi-attention mechanism. It is brave to solve machine reading comprehension tasks in the military field. The model in this paper performs word embedding, part-of-speech embedding, and knowledge base embedding for each word in the article. Among them, knowledge embedding uses military knowledge bases to make up for the lack of information on military entities and relationships in military articles, and independently carries out part-of-speech embedding to solve the problem of part of speech judgment of polysemous words in specific contexts. The experimental results show that the machine reading comprehension model with multi-attention mechanism proposed in this paper can achieve better results on machine reading comprehension tasks, and the accuracy rate is improved by 1.6% compared with ROUGE-L.