ACL(WS)-2022-A sequence-to-sequence approach for document-level relation extraction

[BrambleXu]

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
• johngiorgi/seq2rel + additional community code
• [paper-with-code](

Paper information:

• Author:
• Dataset: BC5CDR, CDR, DocRED, GDA
• keywords:

Notes:

Model Graph:

Result:：

Thoughts:

Next Reading: