Research On The Method Of Reading Comprehension Answer Extraction Based On Discourse Relationship And Graph Structure

Machine reading comprehension is one of the most challenging tasks in the field of natural language processing.With the continuous development of deep learning technology and the release of large-scale Machine reading comprehension data sets,the performance of machine reading and understanding models has constantly refreshed records.However,previous models still have deficiencies in deep semantic understanding and logical reasoning.To alleviate these problems,this paper proposes an answer extraction method for reading comprehension based on discourse relationships and graph structures.The main research contents includes:(1)This paper proposes a recognition method of discourse relations based on elementary discourse units.Discourse relations are analyzed on the basis of semantically independent elementary discourse units.This paper first uses the method based on lexical correlation to obtain the elementary discourse units,and then analyzes the explicit and implicit discourse relations between the elementary discourse units.This method comprehensively considers the context information and semantic factors of the discourse.The experiment shows that the method in this paper can effectively improve the accuracy of the discourse relations model,and provide more knowledge for the next step of reading and understanding answer extraction.(2)A reading comprehension answer extraction method based on CFN and graph structure is proposed.In order to solve the problems of the model in logical reasoning and deep semantic understanding,this paper proposes a reading comprehension answer extraction method based on CFN and graph structure.First,CFN is used to filter out the question-related sentences.Then,the entity relationship diagram is constructed according to the discourse relationship between the discourse units in which the entities are located,the dependency syntax and semantic relationship between the entities by extracting the entities in the text.Finally,the text and the entity relationship graph are integrated into the graph attention network for dynamic logical reasoning to complete the reading logic answer extraction.The paper proposed the method that has been applied to Du Reader-robust dataset,which achieved better results than the baseline model and proved the performance of the reading comprehension answer extraction model improved effectively.This paper proposes a reading comprehension answer extraction method based on discourse relationships and graph structure,which enhances the reading comprehension ability of the model,and puts forward effective solutions for the research of discourse relations between texts that achieves good results in solving machine reading comprehension problems.At the same time,the development of these methods and technologies has also played an excellent driving role in promoting information retrieval,automatic question answering and other fields.