实验结果表明,指针生成器网络显著提升了ROUGE得分,相比基准的序列到序列模型和现有的最优生成式模型,ROUGE-1、ROUGE-2和ROUGE-L得分提高了至少2个点。在加入覆盖机制后,生成摘要中的重复现象几乎被完全消除,重复率与参考摘要相当。此外,模型能够更好地处理源文本中的罕见词汇,并生成准确的事实信息。 生成能力分析 尽...
1.背景及问题描述 seq2seq模型的出现给生成式摘要(abstractive text summarization)的实现提供了更好的方案,但是seq2seq常常出现的两点弊端:1. 它容易生成不准确的细节 2. 容易重复 3.不能处理OOV问题。作者提出了两个方法来提升seq2seq模型的表现,一个是将PointerNetwork与seq2seq中的encoder结合起来,使得生成的结...
GetToThePoint:SummarizationwithPointer-GeneratorNetworksAbigailSeeStanfordUniversityabisee@stanford.eduPeterJ.LiuGoogleBrainpeterjliu@googleChristopherD.ManningStanfordUniversitymanning@stanford.eduAbstractNeuralsequence-to-sequencemodelshaveprovidedaviablenewapproachforab-stractivetextsummarization(meaningtheyarenotrestricte...
代码有三种选项:basic attention model / point generator model / pointer-generator + coverage model 改进都在decoder端,在model.py里是通过self._add_decoder实现的,在这个函数里定义了cell和prev_coverage(只有test的时候才有),再调用封装好的在tf基础上改进的attention_decoder.py函数。 该函数接口为: decoder_...
论文阅读笔记《Get To The Point: Summarization with Pointer-Generator Networks》,程序员大本营,技术文章内容聚合第一站。
【论文阅读】Get to the point :summarization with pointer-generator networks,程序员大本营,技术文章内容聚合第一站。
The traditional approach to automatic summarization is based only on word attention and most of them focus on generating a single sentence summarization. In this work, we propose a novel model with a dual attention that considers both...
pytorch implementation of Get To The Point: Summarization with Pointer-Generator Networks Train with pointer generation and coverage loss enabled Training with pointer generation enabled How to run training Papers using this code Train with pointer generation and coverage loss enabled After training for ...
“Get To The Point: Summarization with Pointer-Generator Networks.” In: arXiv preprint arXiv:1704.04368 Conference Presentation(s): PDF Version: P17-1099.Presentation.pdf Video: https://vimeo.com/234956256 Supplementary Material(s): P17-1099.Notes.pdf Resource(s): https://pypi.org/...
链接:Summarization with Pointer-Generator Networks 介绍: 文本摘要主要有两种形式:抽取式(extractive) 和 生成式(abstractive)。 抽取式摘要是从原文中选取出最重要的几个句子, 重新组织它们以形成摘要, 通常是转化成针对句子的排序或者分类的任务。生成式摘要是利用自然语言生成技术,生成对原文抽象概括性的表述, 主要...