将句子的依存树输入tree-lstm解决医学报告中trigger和argument长距离的问题，将外部KB知识embedding后输与句子的vector和word embedding一同输入lstm
Learning Named Entity Tagger using Domain-Specific Dictionary——AutoNER
Posted on
Edited on
In
Training Data Generation
Views:
Valine:
Symbols count in article: 11k Reading time ≈ 10 mins.
Symbols count in article: 11k Reading time ≈ 10 mins.
为解决标注数据不足问题，通常可以找相似领域的有标记数据做领域迁移、用领域词典做远程监督生成标记数据——本文讨论如何使用领域词表来生成标注数据。提出了一种 Tie or Break 的标注方案
Learning the Structure of Generative Models without Labeled Data
Posted on
Edited on
In
Training Data Generation
,
Generative Models
Views:
Valine:
Symbols count in article: 9.3k Reading time ≈ 8 mins.
Symbols count in article: 9.3k Reading time ≈ 8 mins.
Snorkel related papers 3——优化目标改进，引入稀疏正则，用SGD因此截断梯度法
Data Programming:Creating Large Training Sets, Quickly
Posted on
Edited on
In
Training Data Generation
,
Snorkel
,
Data Programming
,
PGM
,
Factor Graph
,
MC
Views:
Valine:
Symbols count in article: 8.6k Reading time ≈ 8 mins.
Symbols count in article: 8.6k Reading time ≈ 8 mins.
Snorkel related papers 1
Snorkel-- rapid training data creation with weak supervision
Posted on
Edited on
In
Training Data Generation
,
Snorkel
Views:
Valine:
Symbols count in article: 13k Reading time ≈ 12 mins.
Symbols count in article: 13k Reading time ≈ 12 mins.
Snorkel related papers 2——侧重工程实现及演示
Automatically Labeled Data Generation for Large Scale Event Extraction
Posted on
Edited on
In
Event Extraction
,
Distant Supervision
,
Training Data Generation
Views:
Valine:
Symbols count in article: 9.5k Reading time ≈ 9 mins.
Symbols count in article: 9.5k Reading time ≈ 9 mins.
从freebase三元组作为knowledge base进行关系提取的远程监督受到启发迁移到event extraction上，设计了key argument的提取，用于找到trigge和rargument role来确定event type
Event Detection with Neural Networks-- A Rigorous Empirical Evaluation
Posted on
Edited on
In
Event Extraction
,
JRNN related model
Views:
Valine:
Symbols count in article: 4.7k Reading time ≈ 4 mins.
Symbols count in article: 4.7k Reading time ≈ 4 mins.
借助DAG-GRU建模语法信息syntactic information，并对系统性能对模型初始化的敏感性进行了实证研究/div>
Liberal Event Extraction and Event Schema Induction
Posted on
Edited on
In
Event Extraction
Views:
Valine:
Symbols count in article: 5.4k Reading time ≈ 5 mins.
Symbols count in article: 5.4k Reading time ≈ 5 mins.
trigger type自动生成，非分类方式，扩展type集，以distribution hypothesis 为依据，提出两个假设，以这两个假设为依据，将 event trigger 的上下文中包含的 argument和argument对应的role和type 作为给 trigger 进行聚类的依据(通过聚类给trigger生成type，聚的新类就产生新的trigger type，超越ACE这些固有schema和预定义的type)
Event detection and co-reference with minimal supervision
Posted on
Edited on
In
Event Extraction
Views:
Valine:
Symbols count in article: 10k Reading time ≈ 9 mins.
Symbols count in article: 10k Reading time ≈ 9 mins.
event type通过实例自动生成，非分类方式，扩展event type集， 解决共指问题
Open Domain Event Extraction from Twitter
Posted on
Edited on
In
Event Extraction
,
Topic Model
Views:
Valine:
Symbols count in article: 7.3k Reading time ≈ 7 mins.
Symbols count in article: 7.3k Reading time ≈ 7 mins.
从twitter文本中提取calendar event(重要事件)，包括type和确切的时间的pipeline模型