中文新闻文本多文档摘要生成  

Multi-document summary generation of Chinese news text

作　　者：李宝安 佘鑫鹏 常振宁 吕学强[1] 游新冬 LI Bao-an;SHE Xin-peng;CHANG Zhen-ning;LYU Xue-qiang;YOU Xin-dong(Beijing Key Laboratory of Network Culture and Digital Communication,Beijing Information Science and Technology University,Beijing 100101,China)

机构地区：[1]北京信息科技大学网络文化与数字传播北京市重点实验室,北京100101

出　　处：《计算机工程与设计》2023年第9期2867-2873,共7页Computer Engineering and Design

基　　金：国家自然科学基金项目(62171043);北京市自然科学基金项目(4212020);国防科技重点实验室基金项目(6412006200404);北京信息科技大学“勤信人才”培育计划基金项目(QXTCP B201908);北京市市教委科研计划基金项目(KM202111232001)。

摘　　要：针对自动文本摘要任务中多文档生成式摘要模型抽取文本特征不充分、无法捕获跨文档关系和生成内容重复的问题,提出一种多文档生成式摘要模型Transformer-PGN。利用融合段落注意力机制的Transformer结构对文本进行特征抽取,同时捕获跨文档关系;通过指针生成网络逐词生成摘要;结合覆盖率机制避免生成重复单词,形成最终的摘要。实验结果表明,与指针生成网络等模型相比,该模型生成的摘要信息更丰富、冗余度更低,Rouge-1、Rouge-2和Rouge-L分别提升了2.1%、2.4%和2.3%。Aiming at the problems that the multi-document generative summarization model in the automatic text summarization task is insufficient in extracting text features,cannot capture cross-document relationships and generate content duplication,a multi-document generative summarization model Transformer-PGN was proposed.The Transformer structure fused with the paragraph attention mechanism was used to extract features from the text,while capturing cross-document relationships.The pointer generation network was used to generate word-by-word summaries.The coverage mechanism was combined to avoid gene-rating repeated words and form the final summary.Experimental results show that compared with models such as pointer gene-ration network,the summary information generated using this model is richer and less redundant,and Rouge-1,Rouge-2 and Rouge-L are improved by 2.1%,2.4%and 2.3%respectively.

关 键 词：中文新闻 多文档 指针生成网络 文本摘要 深度学习 注意力机制 覆盖率机制 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]