Skip to content

use bert to solve ner task(ucas bigdata homework)

Notifications You must be signed in to change notification settings

zhuyjan/bert-ner

Repository files navigation

UCAS大数据分析编程作业:完成一种基于深度学习的命名实体识别方法

  • base_model使用bert,数据集使用CoNLL-2003
  • results/ 下包括所有模型的收敛曲线
  • 包括bert, bert-crf, bert-cnn, bert-lstm, bert-cnn-lstm模型

开发环境:

  • Python 3.11.10
  • Ubuntu 18.04.1
  • cuda 12.1.105
  • torch2.4.0+cu121
  • NVIDIA GeForce RTX 3090

结果展示:

model&trick accuracy $f_1$ score
bert+softmax 0.9638 0.8223
bert+softmax+余弦退火学习率 0.9668 0.8456
bert+lstm 0.9594 0.8009
bert+cnn 0.9648 0.8128
bert+lstm+cnn 0.9778 0.8606

bert_fuse

声明:

本项目基于https://github.com/xsfmGenius/Ner_Bert_CoNLL-2003 进行开发

About

use bert to solve ner task(ucas bigdata homework)

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages