knowledge-graph-learning
knowledge-graph-learning copied to clipboard
BrambleXu
A curated list of awesome knowledge graph tutorials, projects and communities.
**Summary:** 针对Zero-Shot Relation Triplet Extraction (ZeroRTE) 问题，提出了基于语言模型生成relation的方法 **Resource:** - [pdf](https://aclanthology.org/2022.findings-acl.5/) - declare-lab/relationprompt + [ additional community code](https://paperswithcode.com/paper/?acl=2022.findings-acl.5) - [paper-with-code]( **Paper information:** - Author: - Dataset: [FewRel](https://paperswithcode.com/dataset/fewrel), [Wiki-ZSL](https://paperswithcode.com/dataset/wiki-zsl) - keywords: **Notes:**...
**Summary:** 先行研究已经显示预训练模型里的knowledge augmentation methods是有效的。但是这些方法在不同domain，不同downstream task上的表现是不同的。本研究调查了基于预训练模型和 knowledge graph data在cause-effect relation classification 和 commonsense causal reasoning tasks上的表现。 **Resource:** - [pdf](https://aclanthology.org/2022.csrr-1.6/) - phosseini/causal-reasoning **Paper information:** - Author: - Dataset: [ATOMIC](https://paperswithcode.com/dataset/atomic), [BCOPA-CE](https://paperswithcode.com/dataset/bcopa-ce), [COPA](https://paperswithcode.com/dataset/copa), [GLUCOSE](https://paperswithcode.com/dataset/glucose),...
**Summary:** 本研究没有将distant supervision data 添加到人工标注的训练数据里，而是将基于distant supervision data 的训练模型用于RE的模型。另外也会使用一些标注数据 **Resource:** - [Improving Supervised Drug-Protein Relation Extraction with Distantly Supervised Models](https://aclanthology.org/2022.bionlp-1.16.pdf) - [code]( - [paper-with-code]( **Paper information:** - Author: - Dataset: -...
[RFBFN: A Relation-First Blank Filling Network for Joint Relational Triple Extraction](https://aclanthology.org/2022.acl-srw.2.pdf) **Summary:** 大部分研究忽视了relation和subject, objects的语义信息，本研究提出的模型，首先提取潜在的realtions，然后将lretions转化为relation templates，最后为了填补templates里的空白，找到对应的subject和objects **Resource:** - [pdf](https://aclanthology.org/2022.acl-srw.2/) - lizhe2016/rfbfn - [paper-with-code]( **Paper information:** - Author: - Dataset: NYT* WebNLG*...
**Summary:** 针对跨句子的document-level relation extraction (DocRE)领域的研究。大部分眼前都是pipeline-based，需要将entity作为输入。但是，对entity和relaiton进行联合训练的话，因为共享了参数，所以能有更好的效果。本研究提出的seq2rel，可以学习DocRE的subtask(entity extraction, coreference resolution and relation extraction) . **Resource:** - [A sequence-to-sequence approach for document-level relation extraction](https://aclanthology.org/2022.bionlp-1.2.pdf) - johngiorgi/seq2rel + [ additional community code](https://paperswithcode.com/paper/?acl=2022.bionlp-1.2) - [paper-with-code](...
**Summary:** 本研究将pretrained knowledge base graph embeddings和transformer结合，用于提升sentential Relation Extraction的性能。具体来说，提案的模型基于off-task pretrained graph embeddings和an on-task finetuned BERT encoder。性能不仅很好，而且还观测到了BERT模型的学习行为。 **Resource:** - [Pretrained Knowledge Base Embeddings for improved Sentential Relation Extraction](https://aclanthology.org/2022.acl-srw.29.pdf) - brunoliegibastonliegi/pretrained-kb-embeddings-for-re - [paper-with-code](...
**Summary:** 一片综述型文章。主要着点于cross-dataset, cross-domain。研究显示，标注中的差异会显著影响关系分类的性能，接色儿为什么cross-dataset的评分下降。 **Resource:** - [What Do You Mean by Relation Extraction? A Survey on Datasets and Study on Scientific Relation Classification](https://aclanthology.org/2022.acl-srw.7.pdf) - [Kaleidophon/deep-significance](https://github.com/Kaleidophon/deep-significance) - [paper-with-code]( **Paper information:** - Author:...
[DiS-ReX: A Multilingual Dataset for Distantly Supervised Relation Extraction](https://aclanthology.org/2022.acl-short.95.pdf) **Summary:** Distantly Supervised Relation Extraction这个领域有一个2020年的数据集，RELX，但是因为一些缺点，所以本研究给出了一个新的数据集来做benchmark。 **Resource:** - [pdf](https://aclanthology.org/2022.acl-short.95/) - [dair-iitd/DiS-ReX](https://github.com/dair-iitd/DiS-ReX) - [paper-with-code]( **Paper information:** - Author: - Dataset: [DiS-ReX](https://paperswithcode.com/dataset/dis-rex), [RELX](https://paperswithcode.com/dataset/relx) -...
**Summary:** distantly supervised relation extraction (DS-RE)将每个句子编码为一个entity-pair bag。然后用于bag-level relation prediction。在编码的时候，这种方法不允许信息流入到其他句子的bag里，本研究认为这种情况没有充分利用好bag信息。因此，本研究将一个bag的所有句子连接成一个文章，然后用BERTB编码。然后用relation作为索引，对所有token的contextual embedding进行累计。 **Resource:** - [pdf](https://aclanthology.org/2022.acl-short.38/) - dair-iitd/dsre - [paper-with-code]( **Paper information:** - Author: - Dataset: [DiS-ReX](https://paperswithcode.com/dataset/dis-rex)[DiS-ReX](https://paperswithcode.com/dataset/dis-rex) - keywords: **Notes:** Monolingual DS-RE Multilingual...
**Summary:** 用CNN对entity和relation进行联合训练。本研究的创新点在于将抽取entity和relation当做一个tabel problem，然后用table representation对entity和relation进行建模。将每个talble当做一个image，每个cell当做一个image pixel。 **Resource:** - [pdf](https://aclanthology.org/2022.spnlp-1.2.pdf) - code： youmima/tablert-cnn - [paper-with-code]( **Paper information:** - Author: - Dataset: CoNLL04, ACE05, and ADE - keywords: **Notes:** **Model Graph:** **Result:**： **Thoughts:**...