Research And Design Of Question Answering System Based On Machine Reading Comprehension Technology

Question answering system is a key technology of human-computer interaction,which can read questions and give short answers,and has a wide range of applications.In recent years,thanks to the improvement of the performance of hardware equipment,artificial intelligence technology has the opportunity for rapid development,and has made great progress and has also been widely used.Because of its wide application,many fields have shown the trend of intelligent development.In the field of question answering,more and more people use machine reading comprehension technology to design question answering systems.Machine reading comprehension can read the questions raised by users and documents related to the questions,and give short and precise answers after full understanding.The principle of this question answering system is to first retrieve the most relevant documents from the background data through information retrieval technology,and then use machine reading comprehension technology to read and obtain the answer.It has stronger reasoning ability and does not require complex database design.It is the development direction of intelligent question answering.Therefore,this paper chooses to study it.The effect of the reading comprehension model directly affects the question-answering effect of the system,and the interaction module of the model is the key to the effect of the model.Therefore,aiming at the noise problem in the interaction process,this paper introduces soft thresholding technology into machine reading comprehension,and at the same time uses bidirectional attention information as gating,improves LSTM,and enhances the interaction of information.The built model will be mainly oriented to Chinese question and answer,so the encoding part uses the version provided by the pre-training model BERT for Chinese,and the data set used for the training and testing of the model is the Chinese data set Du Reader.At the same time,this paper will also use this model to implement the question answering system.Knowledge data in the database is accompanied by category information,which is used to narrow the search scope.The retrieval method uses the space vector model and uses the cosine distance to calculate the similarity.The system is also designed with a user feedback function.Users can score the retrieved documents for their relevance,and the collected data can be used to optimize the retrieval model later.This paper also tests the effect of the system through some examples,and manually evaluates the results returned by the system.According to the performance of the system,it analyzes the reasons that affect the quality of the system’s question and answer.