Research And Application Of Machine Reading Comprehension Based On Knowledge Enhancement

In recent years,machine reading comprehension technology is one of the research hotspots in the field of natural language processing.As an important application in the field of natural language processing,intelligent question answering is becoming more and more popular.With the development of deep learning technology and the appearance of high-quality reading comprehension data sets,new models in the field of machine reading comprehension have been proposed and improved constantly,reached a new stage after the emergence of pre-training models.At present,this technology has been widely used in the business industry,such as Taobao,JD and other e-commerce platforms launched e-commerce customer service,Apple launched Siri and so on.However,the existing machine reading comprehension model has some problems in the actual industrial scene,such as the model is not robust,poor mobility,lack of external knowledge utilization,etc.This paper mainly studies how to improve the application of machine reading comprehension model in the actual scene,improve the generalization index of the model and other issues.The main research content is as follows:(1)Aiming at the robustness of current reading comprehension models,a reading comprehension data set enhancement method DA-GPT2 based on GPT2 is proposed.Firstly,the optimal enhancement sentence was found based on the word vector in Glove,and then text was generated by language diversity of GPT-2 pre-training model,and the problem text in the original data set was enhanced.In this way,the training data can be expanded on the whole and the robustness of the model can be improved.(2)To solve the problem of insufficient introduction of external knowledge in the current reading comprehension model,an external knowledge introduction method NANet based on entity recognition and attention mechanism is proposed.Taking the reading comprehension task based on the pre-training model as the main body,an additional auxiliary task is added: entity recognition,and the entity extracted from the auxiliary task interacts with the external knowledge vector based on the attention mechanism,so as to improve the effect of introducing external knowledge into the pre-training model.(3)With the full recovery of the tourism industry brought by the release of the epidemic,intelligent question and answer system has a wide demand in the tourism industry.Based on Rasa,an open source framework,this paper designs and implements an intelligent question and answer system for tourism consulting.In terms of the function of the question and answer system,it implements questions and answers based on FAQ,knowledge graph and machine reading comprehension technology,and realizes the integration of multi-module answers.