Design And Implementation Of Question Answering System Based On Machine Reading Comprehension

With the rapid development of the Internet,society has gradually entered an era of information explosion.The Internet is generating new information all the time,but this information is hidden in other formats such as various free-form texts,ordinary Internet.It is difficult for a user to directly obtain the desired answer from the search results.The emergence of the question and answer system solves this problem.The question and answer system can process the questions raised by the user and then directly give answers,eliminating the process of the user viewing each search result and then summarizing the answer.But the traditional question and answer system has many shortcomings.The traditional question and answer system is mostly based on the rule template.This paper chooses the question and answer system based on machine reading comprehension.The main work is as follows:(1)Propose a two-group based article retrieval model.This paper innovatively uses the two-group word order information as a dictionary,and then uses TF-IDF and other related retrieval techniques to generate an inverted list,and then uses the hash algorithm to improve the retrieval system response time.This model significantly improves the accuracy and efficiency of the search.(2)A more in-depth analysis and understanding of the question and answer system,the word tagging,named entity identification and other sequence tag information as feature input in addition to the semantic word embedding vector,in order to capture more in the question and answer system Information,and can better grasp the focus in shorter questions.(3)A method of classifying the problem and training the model separately is proposed,which improves the performance of the question and answer system.(4)Introducing the attention mechanism and the reading comprehension model of the two-way long-term memory network: This paper introduces the attention mechanism in the two-way long-term memory network to improve the traditional reading comprehension model,and considers the characteristics of different levels to improve the traditional reading.Understanding the model improves the accuracy of the reading comprehension model.This question-and-answer system provides 2% accuracy compared to the DrQAQ&A system and provides a broader perspective for the Q&A system.