Machine Reading Comprehension Based On Question Understanding Enhancement

Machine reading comprehension means that the machine automatically answers the relative questions to the given texts.It is not only a research focus in the field of artificial intelligence and natural language processing,but also a challenging task.Whether the machine understands questions correctly and comprehensively is fundamental for reading comprehension.However,most of the current systems only use attention mechanisms to model questions and do not fully understand questions.In this thesis,we propose the corresponding solutions for this problem.The main works of this thesis are as follows:(1)The reading comprehension method based on multi-dimensional questions understanding is proposed.This thesis improves understanding of questions in the model through the multiple-dimensional method based on the recognition of question types,question important words and incorporation of the external knowledge.Firstly,the Text CNN model is trained get all data types and the question important words are obtained through syntax analysis tree and rules;secondly,the question important words' external knowledge are added;finally,we integrate all information about questions into the reading comprehension model.The experiments on relevant dataset show that: with the integration of multi-dimensional questions understanding,the value of Rouge-L and Bleu-4 of the reading comprehension model increase by about 8.2% and 7.0% respectively.(2)The reading comprehension strategy based on implicit questions without interrogative words is explored.The main idea is to transform implicit questions into explicit questions.First of all,the implicit questions are divided into general implicit question and complex implicit question.Next,for general implicit questions,we add the missing interrogative words to question based on rules.For complex implicit questions,we use the Text RNN model to identify missing interrogative words to question based on the answer and turn it into an explicitquestion.Lastly,we incorporate it into the reading comprehension model.Experiments show that: the metrics of Rouge-L and Bleu-4 of the proposed method increase by 3.6% and 2.0%,respectively;the metrics of Rouge-L and Bleu-4 of the final model increase by 9.5% and 7.8%,respectively.(3)A prototype reading comprehension system for Chinese text is implemented.we implement a reading comprehension system based on the above methods.The system mainly includes five modules:preprocessing,implicit question understanding,question understanding,discourses understanding and answer prediction.Specifically,the questions input to the system are processed in the steps of question understanding,including implicit questions processing,questions classification and questions' important word recognition.Then,the questions and discourses are modeled through attention mechanism.Finally,the system will output the answer.This thesis proposes a reading comprehension method based on multi-dimensional questions understanding for questions enhancement.At the same time,we put forwards the corresponding solving strategies for implicit questions comprehension.Experimental results show that these methods make the model more effective in understanding questions.In the future,we will further strengthen the understanding of important words and abstract words in the model.And we will explore a more effective method to understand implicit questions.