Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[C18-1183] Distantly Supervised NER with Partial Annotation Learning and Reinforcement Learning #2

Open
ftnext opened this issue Apr 12, 2021 · 0 comments

Comments

@ftnext
Copy link
Owner

ftnext commented Apr 12, 2021

https://www.aclweb.org/anthology/C18-1183/
👉 https://aclanthology.org/C18-1183/

#1 でsimple string matchingによるautomatic annotationと紹介されていて興味を持った

時間を決めて気になるところを読んだ


まとめ

中国語のNERで、新しいNER タイプのケース(※)の新しい手法を提案した論文。
(※)人がアノテーションした十分なデータがない

distant supervision: 新しいタイプのエンティティのリストの辞書をもとに、自動でラベル付きのデータを生成する(人手でやるよりも大量にできる)。
ただし2つの問題がある

  1. incomplete annotation: 辞書に含まれないエンティティはアノテーションされない(アノテーション漏れ)
  2. noisy annotation: エンティティのうち一部だけのアノテーション(エンティティ全体がアノテーションされない)

提案手法

  • 1にはpartial annotation(PA)で対応
  • 2は強化学習を使って、正例(positive instance)を選択して対応

Table1, 2を見ると

  • 辞書はprecisionが著しく高いがrecallは著しく低い(正例を逃している)
  • LSTM-CRF(Baseline)にPAとSL(強化学習によるselector)をあわせたアプローチがF1スコア最高

コード

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

1 participant