基于改进Seq2Seq-Attention模型的文本摘要生成方法  

Text summary generation method based on improved Seq2Seq-Attention model

作　　者：门鼎 陈亮[1] MEN Ding;CHEN Liang(School of Computer Science,Xi’an Polytechnic University,Xi’an 710048,China)

机构地区：[1]西安工程大学计算机科学学院,陕西西安710048

出　　处：《电子设计工程》2022年第23期6-10,共5页Electronic Design Engineering

基　　金：陕西省工业攻关项目(2014K05-43);广东省计算机集成制造重点实验室开放基金项目(CIMSOF2016001)。

摘　　要：针对文本摘要生成中词汇语义表达不准确,重复生成以及核心词丢失等问题,提出了一种混合式文本摘要自动生成方法进行摘要抽取,并通过Seq2Seq-Attention模型进行文本摘要生成,在注意力机制中加入核心词,结合指针网络生成模型,该模型可以通过核心词中的重要信息,构建出摘要框架,生成信息全面精炼的文章摘要。解决重复生成以及核心词丢失等问题,文本词汇语义表达准确率明显提高,使得生成的摘要更加流畅。将模型混合进行实验,模型实验数据表明,BERT⁃SUM+Seq2Seq-Attention模型相比于传统模型ROUGE平均值提高了1.6%,混合模型的文本摘要自动生成技术能够提取文本数据中的关键词,形成简单有效的文本段落。Aiming at the problems of inaccurate vocabulary semantic expression,repeated generation and loss of core words in text summary generation,a hybrid text summary automatic generation method is proposed.The BERT model performs abstract sentence extraction,and uses the Seq2Seq⁃Attention model to generate text summaries.The core words are added to the attention mechanism and then combined with the pointer network to generate a model.The model can use the important information in the core words to construct a summary framework.Generate article abstracts with comprehensive and refined information.Solving the problems of repeated generation and loss of core words,the accuracy of the semantic expression of text vocabulary is significantly improved,making the generated abstract more fluent.The model is mixed for experiments.The model experiment data shows that the BERTSUM+Seq2Seq+Attention model is 1.6%higher than the traditional model ROUGE average.The result shows that the text summary automatic generation technology of the hybrid model can extract the keywords in the text data,and the formation is simple A valid text paragraph.

关 键 词：文本摘要 自动生成 Seq2Seq-Attention模型 BERTSUM模型 

分 类 号：TN0[电子电信—物理电子学]