记录一下Pointer Network 相关博客及学习过程,主要记录学习如何使用Pointer Network做抽取式摘要
一、参考博客
https://www.jianshu.com/p/2ad389e91467 详细讲解了一份tf代码
https://blog.csdn.net/qq_20135597/article/details/83586694 翻译了这篇论文的前半部分
https://www.atyun.com/7771.html 一些新鲜的介绍
https://www.jianshu.com/p/bff3a19c59be Ptr-network下NLP领域的使用
https://blog.csdn.net/qq_17827079/article/details/80171728 文章自动摘要之Pointer-Generator Networks的翻译
https://www.sohu.com/a/140364283_297288 (Pointer Network 与)CopyNet 之上有哪些新颖任务
https://mp.weixin.qq.com/s?__biz=MzIxMjAzNDY5Mg==&mid=2650792063&idx=1&sn=fc3b65d6a385edb43102a105aef65c14&scene=0 赛尔原创 | Pointer Networks在自然语言处理领域中的应用
二、在NLP领域中的应用
1)文章摘要
Neural Summarization by Extracting Sentences and Words
Get To The Point: Summarization with Pointer-Generator Networks
2)信息抽取
3)句子排序
三、gayhub代码
3.1 pytorch 实现 Pointer Network
https://github.com/shiretzet/PointerNet
https://github.com/Thidtc/PointerNetwork-PyTorch
3.2 Neural Summarization by Extracting Sentences and Words采用了以下代码的CNN
https://github.com/ambyerhan/pytorch-lstm-char-cnn-master pytorch 实现 CNN-RNN 只有 encoder结构
3.3 Get To The Point: Summarization with Pointer-Generator Networks 的实现代码:
https://github.com/atulkum/pointer_summarizer pointer generation + coverage 和 pointer generation 两种
https://github.com/lipiji/neural-summ-cnndm-pytorch 其中的model2为seq2seq+attention+copy(pointer-network)
3.4 "A Deep Reinforced Model for Abstractive Summarization"的实现代码:
https://github.com/rohithreddy024/Text-Summarizer-Pytorch
四、正式开始
参考的GitHub代码有上述3.1,3.2,3.3
4.1
在Neural Summarization by Extracting Sentences and Words里,sentence是否提取为摘要句的方法为:decoder端隐藏层状态串联encoder端隐藏层状态后接入一个MLP层再接sigmoid激活函数得到句子是否被extract的概率。
*** 打算在一个现有摘要项目架构上进行更改