Design And Implementation Of Knowledge Base Question Answering System Based On Reading Comprehension

With the development of information technology,the explosive growth of Internet information has brought more and more information to Internet users,but it is difficult for ordinary Internet users to quickly locate the correct answers to the problem from the list of web pages retrieved by search engines.In order to meet this demand,the QA answering system has solved the problem of acquiring knowledge quickly,but there are still obvious deficiencies in the current knowledge based QA system.The rapid acquisition of knowledge from the Internet is a very big challenge in the world of computer AI.Different from the traditional QA system,this thesis introduces a model of reading comprehension to find answers to match questions.In order to achieve better results in QA system based on reading comprehension,the main innovations in this thesis are as follows:1.IRGAN-based Relevant Wikipedia Article Retrieval: This thesis combines the knowledge representation learning TransE with the Generative Adversarial Nets retrieval model IRGAN,and designs the score function and the feature construction strategy of IRGAN to improve the accuracy of the Wikipedia document retrieval related to the problem.Compared with One-Hot and Word Embedding,the TransE model enhances the semantic information of free text by using structured knowledge,and provides support for the following IRGAN feature modeling.At the same time,compared with the relevant document retrieval technology based on keyword and TF-IDF,IRGAN generates an obvious improvement in retrieval accuracy.2.Comprehensive Reading Comprehension Model with Attention mechanism: In this thesis,the attention mechanism is introduced into the bidirectional LSTM network to improve the traditional reading comprehension model.After obtain the collection of candidate documents is obtained,and machine reading is needed to obtain the paragraph of the answer.This article creatively introduces the attention mechanism to improve the traditional reading comprehension model,improves the accuracy of the reading comprehension model,and compares the reading comprehension answers to the traditional knowledge questions and answers.It is more readable and can make use of more abundant free text resources.3.The QA system was set up to train and evaluate the algorithms of each part of the question answering system through different data sets,and compared with other algorithm models.