Research On Machine Reading Comprehension Methods Based On Self-attention And Question Decomposition

Machine reading comprehension is a pearl in the field of natural language processing,and it is also the focus of research in this field.It can be widely used in the fields of voice assistant,intelligent customer service,chat robot and so on.In recent years,thanks to the rise of big data and deep learning technology,machine reading comprehension has received extensive attention.There are mainly rule-based methods and deep learning methods in traditional works.The deep learning method encodes the question and context into vector representations,and then uses the attention mechanism to acquire the interactive information between the vector representations.Finally,the semantic interaction information is used to predict the start and end position of the answer.Deep learning has made a breakthrough in the field of machine reading comprehension,but there are still some problems to be solved.In order to solve the problem that it is impossible to distinguish distant similar answers,this thesis proposes a method based on the self-attention mechanism.Self-attention mechanism can ignore the distance between words,directly calculate the dependence between similar answer spans,and then capture the differences between answer spans.In order to solve the complex multi-hop question in the field of reading comprehension,this thesis proposes a machine reading comprehension method based on question decomposition.Firstly,the multi-hop question is decomposed into several single-hop questions,and then the single-hop reading comprehension model is used to solve them.This thesis regards question decomposition as a reading comprehension task:Firstly,the evidence paragraph is used to generate the query,and then the query is used to extract the text spans of single-hop questions from the multi-hop questions.Machine reading comprehension task captures the interactive semantic information between multi-hop questions and evidence paragraphs,which can guide the extraction of single-hop questions in multi-hop questions.The machine reading comprehension method based on self-attention mechanism is tested on the Du Reader dataset,and good results are achieved.This shows that the method based on self-attention is effective.The machine reading comprehension method based on question decomposition has achieved a high score in the Laisi reading comprehension dataset.The experimental results show that the method based on question decomposition can solve the complex multi-hop question.