事件抽取文献整理(2019)

前言

之前研究事件抽取领域(NLP一个小领域信息抽取的子领域), 之前整理过一些文献。

本文是 事件抽取文献整理(2020-2021) 的后续。

事件抽取文献整理(2020-2021)
+
事件抽取文献整理(2019)
+
事件抽取文献整理(2018)
+
事件抽取文献整理(2008-2017)

模型标题后有$代表有给代码

论文

2019

Liu et al. $

Neural Cross-Lingual Event Detection with Minimal Parallel Resources (aclanthology.org)
源码: facebookresearch/MUSE (github.com)

跨语言

Doc2EDAG

Doc2EDAG: An End-to-End Document-level Framework for Chinese Financial Event Extraction (tsinghua.edu.cn)
源码: dolphin-zs/Doc2EDAG (github.com)

用了三个transformer块。

Ananya et al

Cross-lingual Structure Transfer for Relation and Event Extraction (aclanthology.org)
无官方源码
跨语言

Chau et al.

Open-domain Event Extraction and Embedding for Natural Gas Market Prediction (arxiv.org)
源码:
minhtriet/gas_market (github.com)

这个的目的是预测Natural Gas Market

GAIL-ELMo

Joint Entity and Event Extraction with Generative Adversarial Imitation Learning (illinois.edu)
无官方源码

GAN

ODEE-FER $

Open Domain Event Extraction Using Neural Latent Variable Models (aclanthology.org)
源码: lx865712528/ACL2019-ODEE (github.com)

DYGIE++ $

Entity, Relation, and Event Extraction with Contextualized Span Representations (aclanthology.org)
源码: dwadden/dygiepp (github.com)
使用了allenNLP
bert, multi-task, graph

HMEAE $

[HMEAE: Hierarchical Modular Event Argument Extraction (aclanthology.org)](https://aclanthology.org/D19-1584.pdf0
源码：thunlp/HMEAE (github.com) tensorflow
这个仓库里有DMCNN， 然后DMBERT的可见这个仓库 Bakser/DMBERT: A temporary repo to share the DMBERT code for Event Detection (github.com) pytorch

AC 59.3的F1很是让人吃惊~

Han et al.

Joint Event and Temporal Relation Extraction with Shared Representations and Structured Prediction (aclanthology.org)
无官方源码
bert+bilstm

PLMEE

Exploring Pre-trained Language Models for Event Extraction and Generation (aclanthology.org)
有bert

这个不是官方代码: boy56/PLMEE (github.com)， 暂时还没找到官方代码。我看这个boy56/PLMEE的非官方代码使用了AIlenNLP这个库，使得封装更为严重，跟不利于借鉴器模块。

本文解决的是EE问题，提出PLMEE模型，模型由事件抽取模型和生成模型两部分组成，这两个模块都使用到了预训练语言模型来引入更丰富的知识。
根据不同角色对该类型事件的重要性，对损失函数的权重进行了重分配。

本文的EE模型，是先对触发器进行抽取，然后对元素进行抽取，得到元素对应的角色标签。这是一个pipeline的学习过程，损失函数在元素抽取器之后，没有直接对触发器抽取进行优化，可能会出现误差传播问题。

另外，正如作者所说，触发器抽取模块和元素抽取模块，直接利用BERT生成的嵌入表示，没有考虑不同触发器间的关联以及不同元素间的关联。生成模块由于重写adjunct tokens可能会改变原句的语义，因此面临着角色偏离问题。
有一篇读后感的csdn博客：【论文解读 ACL 2019 | PLMEE】Exploring Pre-trained Language Models for Event Extraction and Generation

AEM

Open Event Extraction from Online Text using a Generative Adversarial Network (aclanthology.org)

无官方源码
GAN

JointTransition $

Extracting Entities and Events as a Single Task Using a Transition-Based Neural Model (ijcai.org)
源码: zjcerwin/TransitionEvent (github.com)

bert, bilstm

Li et al.

Biomedical Event Extraction based on Knowledge-driven Tree-LSTM (aclanthology.org)
无官方源码，不太建议搞，还得自己找tree-LSTM代码
Improved Semantic Representations From Tree-Structured Long Short-Term Memory Networks(stanford.edu)
tree-LSTM源码: stanfordnlp/treelstm (github.com) 不过不是pytorch实现的
tree-LSTM

MLM-Joint

Joint Event Extraction Based on Hierarchical Event Schemas From FrameNet | IEEE Journals & Magazine | IEEE Xplore

给了Weili-NLP/EventSchemasBasedOnFrameNet (github.com)， 但无源码

Joint3EE

One for all: Neural joint modeling of entities and events (arXiv)
无官方源码

Chan et al. $

Rapid Customization for Event Extraction (aclanthology.org)
Our system (code, UI, documentation, demonstration video) is released as open source.1
源码: BBN-E/Rapid-customization-events-acl19 (github.com)

Position embeddings (PE):
$P{E}_t$ encodes the relative distance of each word to the trigger word.
$P{E}_a$ to encode relative distances to the candidate argument.