Event extraction and documentation (2020-2021)

Preface

Previously, I studied the field of event extraction (NLP A sub domain of small domain information extraction ), I have collated some documents before .

Event extraction and document sorting (2020-2021)
Event extraction and document sorting (2018-2019)

Model review
The picture is from : A Compact Survey on Event Extraction: Approaches and Applications

I read this description before NLP Event extraction Overview （ in ）—— Model

In the model $ For code

The paper

2021

Gen-arg $

Document-Level Event Argument Extraction by Conditional Generation (aclanthology.org)



Used Bart Model , But I feel incomplete after reading the official source code

BRAD

Event Extraction from Historical Texts: A New Dataset for Black Rebellions (aclanthology.org)
No official source code
A new data set is proposed ( The paper does not give links to public datasets ), Is the main contribution of this article .
a corpus of nineteenth-century African American newspapers.
Our dataset features 5 entity types, 12 event types, and 6 argument roles that concern slavery and black movements between the eighteenth and nineteenth centuries.

TEXT2EVENT $

original text : https://aclanthology.org/2021.acl-long.217.pdf
Code : luyaojie/Text2Event (github.com)
How to integrate shcema Go to constraint decode The process may refer to

CasEE $

CasEE: A Joint Learning Framework with Cascade Decoding for Overlapping Event Extraction (aclanthology.org)
Code : JiaweiSheng/CasEE: Source code for ACL 2021 finding paper: CasEE: A Joint Learning Framework with Cascade Decoding for Overlapping Event Extraction (github.com)
Chinese oriented
I tried the environment , Found no problem , Can run
Also a simple look at the code

This article is actually a reference CasRel (arxiv.org), A triple relation extraction task . Migrate this paradigm to event extraction .

CasEE framework :

Used CLN(Conditioned LayNorm) and MSA(multiHead Self-Attention)

With two pointers , start pos, end pos, But the disadvantage is that the threshold needs to be set manually We select tokens with The mathematical formula : ${\hat{t}}_i^{sc}>{\xi }_2$ as the start positions, and those with The mathematical formula : ${\hat{t}}_i^{ec}>{\xi }_3$as end positions, where The mathematical formula : ${\xi }_2,{\xi }_3\in [0,1]$ are scalar thresholds.

When categorizing arguments , There's another. type_soft_constrain The operation of

p_s = torch.sigmoid(self.head_cls(inp))  # [b, t, l]
p_e = torch.sigmoid(self.tail_cls(inp))

type_soft_constrain = torch.sigmoid(self.gate_linear(type_emb))  # [b, l]
type_soft_constrain = type_soft_constrain.unsqueeze(1).expand_as(p_s)
p_s = p_s * type_soft_constrain
p_e = p_e * type_soft_constrain

Different models have different learning rates , in addition get_cosine_schedule_with_warmup You can see this example : Sentiment analysis bert family pytorch Realization (ing)

def set_learning_setting(self, config, train_loader, dev_loader, model):
        instances_num = len(train_loader.dataset)
        train_steps = int(instances_num * config.epochs_num / config.batch_size) + 1

        print("Batch size: ", config.batch_size)
        print("The number of training instances:", instances_num)
        print("The number of evaluating instances:", len(dev_loader.dataset))

        bert_params = list(map(id, model.bert.parameters()))

        other_params = filter(lambda p: id(p) not in bert_params, model.parameters())
        optimizer_grouped_parameters = [{
    'params': model.bert.parameters()}, {
    'params': other_params, 'lr': config.lr_task}]

        optimizer = AdamW(optimizer_grouped_parameters, lr=config.lr_bert, correct_bias=False)
        scheduler = get_linear_schedule_with_warmup(optimizer, num_warmup_steps=train_steps * config.warmup, num_training_steps=train_steps)

CLEVE $

CLEVE: Contrastive Pre-training for Event Extraction (aclanthology.org)
Code : THU-KEG/CLEVE (github.com)

BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding
Graph Isomorphism Network
Here we use a state-of-the-art GNN model, Graph Isomorphism Network (Xu et al., 2019), as our graph encoder for its strong representation ability.

FEAE

Trigger is Not Sufficient: Exploiting Frame-aware Knowledge for Implicit Event Argument Extraction (aclanthology.org)
No official source code

MRC-based Argument Extraction
Teacher-student Framework

GIT $

Document-level Event Extraction via Heterogeneous Graph-based Interaction Model with a Tracker (aclanthology.org)
Source code : Document-level Event Extraction via Heterogeneous Graph-based Interaction Model with a Tracker (aclanthology.org)

The author in AI Drive Share GIT Video sharing of It is also said in , At first, it was not end to end Trained , But first gold label, Slowly replace with the output of the model

tracker Non parallelism becomes the bottleneck of model running speed , in addition , The order of argument extraction needs to be defined in advance
Like here Equity Freeze It needs to be defined manually Equity Holder -> FrozeShare -> StartDate…
This requires training to find good or bad

github It's a financial data set

NoFPFN $

Revisiting the Evaluation of End-to-end Event Extraction (aclanthology.org)
Source code : dolphin-zs/Doc2EDAG (github.com)

reinforcement learning, to support diverse preferences of evaluation metrics motivated by different scenarios, we propose a new training paradigm based on reinforcement learning for a typical end-to-end EE model,

GATE $

GATE: Graph Attention Transformer Encoder for Cross-lingual Relation and Event Extraction (arxiv.org)

Ahmad, W. U., Peng, N., & Chang, K.-W. (2021). GATE: Graph Attention Transformer Encoder for Cross-lingual Relation and Event Extraction. Proceedings of the AAAI Conference on Artificial Intelligence, 35(14), 12462-12470. Retrieved from https://ojs.aaai.org/index.php/AAAI/article/view/17478

Source code :wasiahmad/GATE (github.com)
Cross language

DualQA

What the Role is vs. What Plays the Role: Semi-Supervised Event Argument Extraction via Dual Question Answering | Proceedings of the AAAI Conference on Artificial Intelligence
No official source code

GRIT $

GRIT: Generative Role-filler Transformers for Document-level Event Entity Extraction (aclanthology.org)

Source code : xinyadu/grit_doc_event_entity (github.com)

Event Entity Extraction

Partially causal masking strategy

Wen et al.

Event Time Extraction and Propagation via Graph Attention Networks (aclanthology.org)
No official source code

2020

SciBERT $

Biomedical Event Extraction as Multi-turn Question Answering (aclanthology.org)
Source code :allenai/scibert: A BERT model for scientific text. (github.com)

Biomedical event extraction
describing specific relationships between multiple molecular entities, such as genes, proteins, or cellular components

Visualization tools BioNLP Shared Task 2011: Supporting Resources (aclanthology.org)

HPNet

Joint Event Extraction with Hierarchical Policy Network (aclanthology.org)
No official source code

M2E2

Cross-media Structured Common Space for Multimedia Event Extraction (aclanthology.org)
No official source code

MQAEE

Event Extraction as Multi-turn Question Answering (aclanthology.org)
No official source code

Model structure :

Du et al. $

Event Extraction by Answering (Almost) Natural Questions (aclanthology.org)
Source code : xinyadu/eeqa: Event Extraction by Answering (Almost) Natural Questions (github.com)
ll Students are using this , Put it first , See what he says

Min et al.

Towards Few-Shot Event Mention Retrieval: An Evaluation Framework and A Siamese Network Approach (aclanthology.org)
No official source code

• Sample pairs that are both in the query, and assign them the same class label.
• Sample pairs such that one of them is in the query but the other is not, and assign this pair the not in same class label.

Chen et al.

Reading the Manual: Event Extraction as Definition Comprehension (aclanthology.org)
No official source code

It can mainly face zero samples and few samples
I don't understand yet Approach part …
trigger cls: 72.9
arg cls: 42.4

EEGCN $

Edge-Enhanced Graph Convolution Networks for Event Detection with Syntactic Relation (aclanthology.org)

Source code : cuishiyao96/eegcned (github.com)

The graph model is a little strange
Edge-Aware Node Update Module first aggregates information from neighbors of each node through specific edge, and Node-Aware Edge Update module refines the edge representation with its connected nodes.

Only events trigger Mission