[EMNLP2017]End-to-End Neural Relation Extraction with Global Optimization
来源:互联网 发布:对网络课程的认识 编辑:程序博客网 时间:2024/06/06 02:10
我们首先了解该paper解决的问题,也就是任务:
句子级的实体识别和关系判断,针对Figure 1给出的一句话:Associated Press writer Patrick McDowell in Kuwait City 识别这句话中出现的识别,其实就是给句子打标签,在本在本文中,实体的标签表示如下:
B-
在识别完实体之后,针对识别出的实体进行两两关系判断,关系的标签类型如下:
下图给出了本文在做实体识别和关系抽取时最主要的流程:
仔细观察Figure 2的table的上半三角中每个表格,每个表格中都有数字,就是解析过程中每一步.表格的个数是
在解析的过程中,仍然是识别句子中所有的entity, 就是首先填充对角线,然后针对识别出的实体进行两两判断关系,但是本文用了beam search,在每一步都保留之前解析结果的top k个得分最好的,第i+1步时保留的top k个解析结果的排序已经发生了变化,而且重要的是,将实体识别序列和关系识别序列一起打分的,所以说是全局最优的.
打分函数如下:
部分填充的table:
根据公式(1)那么目前的问题是:如何用embedding表示
在上图的(a), (b)中的实体识别和关系判定的神经网络中都有一个
一个解析序列的得分就是
在选取特征时用了LSTM-Minus, 我觉得挺有意思的.
- [EMNLP2017]End-to-End Neural Relation Extraction with Global Optimization
- [ACL2016]End-to-end Relation Extraction using LSTMs on Sequence and Tree Structures
- 知识图谱4-【再看一篇论文《End-to-End Relation Extraction using LSTMs on Sequences and Tree Structures》】
- [EMNLP2017]Adversarial Training for Relation Extraction
- 【论文笔记】Neural Relation Extraction with Multi-lingual Attention
- An End-to-End System for Unconstrained Face Verification with Deep Convolutional Neural Networks
- Towards End-to-end Text Spotting with Convolutional Recurrent Neural Networks阅读笔记
- 车牌识别“Towards End-to-End Car License Plates Detection and Recognition with Deep Neural Networks”
- An End-to-End Trainable Neural Network...笔记
- 【论文笔记】An End-to-End Model for QA over KBs with Cross-Attention Combining Global Knowledge
- Explaining How a Deep Neural Network Trained with End-to-End Learning Steers a Car论文笔记
- 车牌检测识别--Towards End-to-End Car License Plates Detection and Recognition with Deep Neural Networks
- [论文阅读笔记]Neural Relation Extraction with Selective Attention over Instances
- 知识图谱5-【继续看论文《Neural Relation Extraction with Selective Attention over Instances》】
- End-to-End Answer Chunk Extraction and Ranking for Reading Comprehension
- end-to-end 理解
- [EMNLP2017]Context-Aware Representations for Knowledge Base Relation Extraction(short paper))
- DenseBox: Unifying Landmark Localization with End to End Object Detection
- viewer.js插件简单使用说明
- 安卓学习笔记(实战)
- android studio 快速集成腾讯bugly
- 评测指标—准确率(Precision)、召回率(Recall)、F值(F-Measure)
- Go学习笔记(二)
- [EMNLP2017]End-to-End Neural Relation Extraction with Global Optimization
- Unity3: DoTween插件的简单使用
- leetcode 42|407. Trapping Rain Water 1|2
- ThinkingInJava
- Jackson 中JSON的三种处理方式
- linux下ftp服务器的安装
- Java中的参数传递
- Sublime Text 之运行 js 方法
- 耳机煲机攻略