Research On Machine Reading Comprehension Of Fragment Extraction Based On Deep Learning

This article mainly discusses the machine reading comprehension task of extracting answers in Chinese fragment type,and extracts the answers in a certain way according to the given questions and related text fragments.The research of machine reading comprehension is one of the directions of the future intelligent question answering system.It has a wide range of applications and has important research significance.In recent years,with the development of deep learning,research on end-to-end reading comprehension models has developed to a certain height.The machine reading comprehension model is generally composed of a text presentation layer,a coding layer,a problem and text interaction layer,and an output layer.It is necessary to extract the feature information of the text from different aspects in order to better understand the problem and the relevant text fragment information.Since the introduction of the pre-trained language model BERT,its performance on various natural language processing tasks exceeds that of previous deep learning models.One of the tasks is machine reading comprehension.The main idea of the reading comprehension model based on the pre-trained language model BERT is to first use a deep learning network to unsupervisedly learn the text representation,and then deploy it on downstream machine reading comprehension tasks through fine-tuning.This article mainly uses the QANet model and the machine reading comprehension model based on the recently proposed pre-trained language model BERT to experiment on three different Chinese reading comprehension data sets: Cail2019,CMRC 2018,and DRCD.The experiment proves that the machine reading comprehension model based on the pre-trained language model BERT has a stronger ability to understand text information,which is reflected in the evaluation index of machine reading comprehension.The accuracy EM and F1 values far exceed the performance of the QANet model.EM value increased by an average of 15.74%,F1 value increased by an average of 9.21%.