Oriented To The Study Of The Discourse Relationship Between Chinese Reading Comprehension

With the transfer of research objects in the field of natural language processing,many researchers focus on the research of larger semantic levels such as sentences and fragments,focusing on understanding the overall structure and semantic association of internal components of text fragments,and as a basic task such as machine translation,question answering,event extraction,etc.,discourse analysis is particularly important.As the basic task of discourse analysis,the study of discourse relationship has gradually become a research hotspot in recent years.The study of discourse relationship aims to study the semantic relationship between elements(sentences,compound sentences,sentence groups,etc.)within texts,and to carry out a series of studies on this.The relationship between discourse is the basic research of the task of discourse analysis,and it is also the key to discourse analysis.In addition,discourse relationships guide reading comprehension tasks,but there is currently a lack of relevant datasets.Therefore,this paper focuses on the relevant corpus of reading comprehension to study the relationship between texts.The research content of this paper mainly includes the following three aspects:(1)Construction of reading comprehension discourse relationship datasetAiming at the problem of lack of Chinese reading comprehension discourse relationship dataset,this paper constructs a reading comprehension discourse relationship dataset.Based on the semantic system of Chinese discourse relationships,this paper uses the natural language processing toolkit to perform sentence segmentation processing on reading comprehension discourse,and then identifies discourse relationships,determines the annotated information(including elements,semantic types of discourse relationships,explicit or implicit discourse relationships,etc.),and formulates corresponding annotation processes to ensure the accuracy of the labeled datasets and the efficiency of annotation.In this paper,331 reading comprehension discourses are labeled,and there are 5996 types of discourse relationships,and the Kappa value of the annotation results is greater than 0.7,indicating the validity of the dataset constructed in this paper.(2)The Implicit Argument Labeling Based on Bipartite Graph NetworkChinese implicit argument labeling aims to identify the boundaries between arguments without explicit connectives.At present,most of the relevant research on argument labeling focuses on explicit argument labeling.Therefore,this paper focuses on implicit argument labeling,and proposes an implicit argument labeling method based on bipartite graph network.This paper uses the method of named entity recognition to study the argument labeling.Firstly,the argument is encoded by the pre-trained language model,and then the association features of each part of the argument element are trained by the bipartite graph neural network,and finally the label sequence of the argument is obtained through the conditional random field.In this paper,experiments are carried out on a self-built dataset,and the results show that the F1 value of the argument recognition is higher than that of the baseline model,which verifies the effectiveness of the proposed method.