Teaching Machines to Read and Comprehend
来源:互联网 发布:php获取手机号码 编辑:程序博客网 时间:2024/06/07 17:33
关键词
real natural language traning data, nerual model
来源
Teaching Machines to Read and Comprehend
arXiv 2015.06.10 (published at NIPS 2015)
问题
针对阅读理解缺乏大规模训练数据集,从CNN和Daily Mail获取数据,构建了相应的数据集。文章直接做document,关键点和总结用来做query。为了方便数据使用,将人名全部替换为”ent123”类似的样子。然后尝试利用神经网络模型解决机器阅读理解问题。
要解决的问题
1.片段主义分析(Frame-Semantic Parsing)
即判断“who did what to him”。
2.词距判断(word distance Benchmark)
使用模型
Deep LSTM Reader
该模型重新设计LSTM公式,如下:
其中的的”||”表示连接两个向量,”|||”代表query和document的分隔符
Attentive Reader
u(query)由最后一个正向lstm输出和最后一个逆向lstm输出拼接而成。
r(document)的计算公式如下:
r是y*s的和累积。
g的计算公式如下:
Impatient Reader
u(query)由最后一个正向lstm输出和最后一个逆向lstm输出拼接而成。
r(document)计算公式如下:
与Attentive Reader区别是,每个query词都算一个r。
g计算公式如下:
实验结果及结论
文章提供新的较大的数据集,并且指出 CNN 语料要比 Daily Mail 阅读理解难度要低一些。
阅读全文
0 0
- Teaching Machines to Read and Comprehend
- Teaching Machines to Read and Comprehend
- 论文“Teaching Machines to Converse”阅读笔记
- Teaching Machines to Converse--阅读笔记1 摘要及介绍
- Teaching John The Ripper how to Crack MD5 Hashes and more
- Teaching Mario to play with himself: AI, machine learning, and Super Mario Bros.
- State Machines and Automata
- Configuring Servers and Machines
- Teaching two russian how to use Oracle.
- Intend to read papers and books
- read from and write to file
- How To Convert Virtual Machines Between VirtualBox and VMware[测试可行]
- Step by step guide to set up master and slave machines
- Facebook AI Director Yann LeCun on His Quest to Unleash Deep Learning and Make Machines Smarter
- Is it OK to mix categorical and continuous data for SVM (Support Vector Machines)?
- BasicExcel - A Class to Read and Write to Microsoft Excel
- Virtual Machines and Hypervisors: Introduction
- Say goodbye to virtual machines
- 网络判断,简单的没网调取缓存图片
- linux工作队列
- tuple index out of range
- springMVC--秒杀系统总结
- 判断一个单链表是否存在环
- Teaching Machines to Read and Comprehend
- 业务逻辑.pdf
- Day012
- 巨人养成
- 实验二运算符的重载,体验深拷贝
- 不要欺负我不懂java的volatile
- java HashSet / TreeSet 学习
- SVM分类器实践,检测是否有篮球
- 爬取百度词条内链接