BETES:一种中文长文档抽取式摘要方法

Method of Extractive Summarization Chinese Long Documents

在自然语言处理领域是最重要的研究工作之一,并随着深度学习的兴起成为研究热点,而中文长文本的摘要抽取面临更大的挑战,存在长文本-摘要语料库不足、摘要抽取信息不准确、目标摘要冗余、摘要句缺失等问题.本文以中文长文本的摘要抽取为研究对象,提出一种BETES方法,基于规则和人工辅助筛选构建中文长文本-摘要语料库;利用Bert预处理模型进行文本向量化,更好地捕捉长文本上下文的语义,提升信息抽取的准确性;在识别中文长文本的基本篇章单元的基础上,以基本篇章单元为抽取对象,降低摘要抽取的冗余度;最后利用Transformer神经网络抽取模型,实现基本篇章单元的抽取,提升摘要句抽取的准确率.实验证明,提出的BETES方法在中文长文本的抽取式摘要过程中提高了准确性,降低了冗余度,并且ROUGE分数优于主流的摘要抽取方法.

Text summarization is one of the most important researches in the field of natural language processing and has become a research hotspot with the rise of deep learning, however, extractive summarization of Chinese long text faces greater challenges, it has some problems, such as insufficient long document summary corpus, inaccurate extraction information, redundant target summary and missing summary sentences.Extractive summarization of Chinese long text is the research object in this paper, we propose a BETES method, To construct Chinese long text-abstract corpus based on rules and manual assisted filtering;The Bert preprocessing model is used for text vectorization to better capture the semantics of long text context and improve the accuracy of information extraction;On the basis of recognizing the Elementary Discourse Units(Edu)of Chinese long text, Taking the Edu as the extraction object, reduce the redundancy of summarization;Finally, Transformer neural network extraction model is used to realize the extraction of Edus and improve the accuracy of summarization sentence extraction.Experiments show that the proposed BETES method can improve the accuracy and reduce the redundancy in the process of extracting long Chinese text, and the ROUGE score is superior to the mainstream summarization extraction method.