Relation enhanced sentence sorting method based on Bert model

Abstract

The invention discloses a relation-enhanced sentence sorting method based on a Bert model, which utilizes the Bert model to mine semantic and logical relations among sentences and performs more comprehensive modeling on the coherence of texts. According to the invention, the method comprises the steps: establishing a relation enhancement type pointer network decoder, and efficiently integrating the reliable relative sequence information into a pointer network through a relation enhancement module; enabling the relation enhancement module to accurately capture the relative sequence of sentencepairs and construct high-level semantic representation of the sentence pairs based on a Bert model, so as to help a pointer network to mine the global relation between unordered sentences and fully consider the local dependence between current candidate sentences and sentences of which the sequence is already arranged; therefore, the network can obtain local and global coherence of the text at thesame time, and a smoother and more reasonable sentence sequence is generated. Compared with a previous sorting method, the technology provided by the invention has the advantages that the sentence sorting accuracy is remarkably improved, and the practical value is relatively good.

Description

technical field [0001] The invention belongs to the technical field of natural language processing, and relates to a relationship-enhanced sentence sorting method based on a Bert model. Background technique [0002] Modeling the coherence of text is one of the key problems in natural language processing. A well-organized and logical text is easier for people to read and understand; on the contrary, incorrect sentence order will lead to ambiguity in the content of the text and reduce readability. The sentence sorting task plays a very important role in enhancing the coherence of the text. This task is to reorder some sentences whose order is disordered or unknown by analyzing the semantic relationship and logical information between the sentences, so that they can form a smooth and coherent structure. of the text. Sentence sorting tasks have a wide range of applications in the field of natural language processing, such as automatic summarization of multiple documents, compo...

AI Technical Summary

Problems solved by technology

However, the disadvantage of these works is that they do not pay enough attention to the local information of the text, and use the relative order information to enhance the entire network at the same time, including the encoder and decoder.

In addition, the current work lacks a better relative order prediction method. They calculate the relative information of sentences based on the sentence vectors of two sentences, and do not directly conn...

Images