Chinese named entity recognization (bert/roberta/macbert/bert_wwm with Keras)

Related tags

Text Data & NLPBERT-NER-Keras
Overview

Chinese named entity recognization (bert/roberta/macbert/bert_wwm with Keras)

Project Structure

./
├── DataProcess
│   ├── __pycache__
│   ├── convert2bio.py
│   ├── convert_jsonl.py
│   ├── handle_numbers.py
│   ├── load_data.py
│   └── statistic.py
├── README.md
├── __pycache__
├── chinese_L-12_H-768_A-12                                    BERT权重
│   ├── bert_config.json
│   ├── bert_model.ckpt.data-00000-of-00001
│   ├── bert_model.ckpt.index
│   ├── bert_model.ckpt.meta
│   └── vocab.txt
├── chinese_bert_wwm                                           BERT_wwm权重
│   ├── bert_config.json
│   ├── bert_model.ckpt.data-00000-of-00001
│   ├── bert_model.ckpt.index
│   ├── bert_model.ckpt.meta
│   └── vocab.txt
├── chinese_macbert_base                                       macBERT权重
│   ├── chinese_macbert_base.ckpt.data-00000-of-00001
│   ├── chinese_macbert_base.ckpt.index
│   ├── chinese_macbert_base.ckpt.meta
│   ├── macbert_base_config.json
│   └── vocab.txt
├── chinese_roberta_wwm_ext_L-12_H-768_A-12                    roberta权重
│   ├── bert_config.json
│   ├── bert_model.ckpt.data-00000-of-00001
│   ├── bert_model.ckpt.index
│   ├── bert_model.ckpt.meta
│   └── vocab.txt
├── config                                                     
│   ├── __pycache__
│   ├── config.py                                              配置文件
│   └── pulmonary_label2id.json                                label id
├── data                                                       数据集
│   ├── pulmonary.test
│   ├── pulmonary.train
│   └── sict_train.txt
├── environment.yaml                                           conda环境配置文件
├── evaluate.py
├── generator_train.py
├── keras_bert                                                 keras_bert（可pip下）
├── keras_contrib                                              keras_contrib（可pip下）
├── log                                                        训练nohup日志
│   ├── chinese_L-12_H-768_A-12.out
│   ├── chinese_macbert_base.out
│   ├── chinese_roberta_wwm_ext_L-12_H-768_A-12.out
│   └── electra_180g_base.out
├── model.py                                                   模型构建文件
├── models                                                     保存的模型权重
│   ├── pulmonary_chinese_L-12_H-768_A-12_ner.h5
│   ├── pulmonary_chinese_bert_wwm_ner.h5
│   ├── pulmonary_chinese_macbert_base_ner.h5
│   └── pulmonary_chinese_roberta_wwm_ext_L-12_H-768_A-12_ner.h5
├── predict.py                                                 预测
├── report                                                     模型实体F1评估报告
│   ├── pulmonary_chinese_L-12_H-768_A-12_evaluate.txt
│   ├── pulmonary_chinese_L-12_H-768_A-12_predict.json
│   ├── pulmonary_chinese_bert_wwm_evaluate.txt
│   ├── pulmonary_chinese_bert_wwm_predict.json
│   ├── pulmonary_chinese_macbert_base_evaluate.txt
│   ├── pulmonary_chinese_macbert_base_predict.json
│   ├── pulmonary_chinese_roberta_wwm_ext_L-12_H-768_A-12_evaluate.txt
│   └── pulmonary_chinese_roberta_wwm_ext_L-12_H-768_A-12_predict.json
├── requirements.txt                                           pip环境
├── test.py                                                    
├── train.py                                                   训练
└── utils                                                      
    ├── FGM.py                                                 FGM对抗
    ├── __pycache__
    └── path.py                                                所有路径

56 directories, 193 files

Dataset

三甲医院肺结节数据集，20000+字，BIO格式，形如：

中	B-ORG
共	I-ORG
中	I-ORG
央	I-ORG
致	O
中	B-ORG
国	I-ORG
致	I-ORG
公	I-ORG
党	I-ORG
十	I-ORG
一	I-ORG
大	I-ORG
的	O
贺	O
词	O

ATTENTION: 在处理自己数据集的时候需要注意：

  • 字与标签之间用空格（"\ "）隔开
  • 其中句子与句子之间使用空行隔开

Steps

  1. 替换数据集
  2. 使用DataProcess/load_data.py生成label2id.txt文件
  3. 修改config/config.py中的MAX_SEQ_LEN（超过截断，少于填充，最好设置训练集、测试集中最长句子作为MAX_SEQ_LEN）
  4. 下载权重，放到项目中
  5. 修改public/path.py中的地址
  6. 根据需要修改model.py模型结构
  7. 修改config/config.py的参数
  8. 训练前debug看下input_train_labels,result_train对不对，input_train_types全是0
  9. 训练

Model

BERT

roberta

macBERT

BERT_wwm

Train

运行train.py

Evaluate

运行evaluate/f1_score.py

BERT

           precision    recall  f1-score   support

     SIGN     0.6651    0.7354    0.6985       189
  ANATOMY     0.8333    0.8409    0.8371       220
 DIAMETER     1.0000    1.0000    1.0000        16
  DISEASE     0.4915    0.6744    0.5686        43
 QUANTITY     0.8837    0.9157    0.8994        83
TREATMENT     0.3571    0.5556    0.4348         9
  DENSITY     1.0000    1.0000    1.0000         8
    ORGAN     0.4500    0.6923    0.5455        13
LUNGFIELD     1.0000    0.5000    0.6667         6
    SHAPE     0.5714    0.5714    0.5714         7
   NATURE     1.0000    1.0000    1.0000         6
 BOUNDARY     1.0000    0.6250    0.7692         8
   MARGIN     0.8333    0.8333    0.8333         6
  TEXTURE     1.0000    0.8571    0.9231         7

micro avg     0.7436    0.7987    0.7702       621
macro avg     0.7610    0.7987    0.7760       621

roberta

           precision    recall  f1-score   support

  ANATOMY     0.8624    0.8545    0.8584       220
  DENSITY     0.8000    1.0000    0.8889         8
     SIGN     0.7347    0.7619    0.7481       189
 QUANTITY     0.8977    0.9518    0.9240        83
  DISEASE     0.5690    0.7674    0.6535        43
 DIAMETER     1.0000    1.0000    1.0000        16
TREATMENT     0.3333    0.5556    0.4167         9
 BOUNDARY     1.0000    0.6250    0.7692         8
LUNGFIELD     1.0000    0.6667    0.8000         6
   MARGIN     0.8333    0.8333    0.8333         6
  TEXTURE     1.0000    0.8571    0.9231         7
    SHAPE     0.5714    0.5714    0.5714         7
   NATURE     1.0000    1.0000    1.0000         6
    ORGAN     0.6250    0.7692    0.6897        13

micro avg     0.7880    0.8261    0.8066       621
macro avg     0.8005    0.8261    0.8104       621

macBERT

           precision    recall  f1-score   support

  ANATOMY     0.8773    0.8773    0.8773       220
     SIGN     0.6538    0.7196    0.6851       189
  DISEASE     0.5893    0.7674    0.6667        43
 QUANTITY     0.9070    0.9398    0.9231        83
    ORGAN     0.5882    0.7692    0.6667        13
  TEXTURE     1.0000    0.8571    0.9231         7
 DIAMETER     1.0000    1.0000    1.0000        16
TREATMENT     0.3750    0.6667    0.4800         9
LUNGFIELD     1.0000    0.5000    0.6667         6
    SHAPE     0.4286    0.4286    0.4286         7
   NATURE     1.0000    1.0000    1.0000         6
  DENSITY     1.0000    1.0000    1.0000         8
 BOUNDARY     1.0000    0.6250    0.7692         8
   MARGIN     0.8333    0.8333    0.8333         6

micro avg     0.7697    0.8180    0.7931       621
macro avg     0.7846    0.8180    0.7977       621

BERT_wwm

           precision    recall  f1-score   support

  DISEASE     0.5667    0.7907    0.6602        43
  ANATOMY     0.8676    0.8636    0.8656       220
 QUANTITY     0.8966    0.9398    0.9176        83
     SIGN     0.7358    0.7513    0.7435       189
LUNGFIELD     1.0000    0.6667    0.8000         6
TREATMENT     0.3571    0.5556    0.4348         9
 DIAMETER     0.9375    0.9375    0.9375        16
 BOUNDARY     1.0000    0.6250    0.7692         8
  TEXTURE     1.0000    0.8571    0.9231         7
   MARGIN     0.8333    0.8333    0.8333         6
    ORGAN     0.5882    0.7692    0.6667        13
  DENSITY     1.0000    1.0000    1.0000         8
   NATURE     1.0000    1.0000    1.0000         6
    SHAPE     0.5000    0.5714    0.5333         7

micro avg     0.7889    0.8245    0.8063       621
macro avg     0.8020    0.8245    0.8104       621

Predict

运行predict/predict_bio.py

Owner
GitHub Repository
PORORO: Platform Of neuRal mOdels for natuRal language prOcessing

PORORO: Platform Of neuRal mOdels for natuRal language prOcessing pororo performs Natural Language Processing and Speech-related tasks. It is easy to

Kakao Brain 1.2k Dec 21, 2022
🤗Transformers: State-of-the-art Natural Language Processing for Pytorch and TensorFlow 2.0.

State-of-the-art Natural Language Processing for PyTorch and TensorFlow 2.0 🤗 Transformers provides thousands of pretrained models to perform tasks o

Hugging Face 77.3k Jan 03, 2023
This project uses word frequency and Term Frequency-Inverse Document Frequency to summarize a text.

Text Summarizer This project uses word frequency and Term Frequency-Inverse Document Frequency to summarize a text. Team Members This mini-project was

1 Nov 16, 2021
Download videos from YouTube/Twitch/Twitter right in the Windows Explorer, without installing any shady shareware apps

youtube-dl and ffmpeg Windows Explorer Integration Download videos from YouTube/Twitch/Twitter and more (any platform that is supported by youtube-dl)

Wolfgang 226 Dec 30, 2022
TEACh is a dataset of human-human interactive dialogues to complete tasks in a simulated household environment.

TEACh is a dataset of human-human interactive dialogues to complete tasks in a simulated household environment.

Alexa 98 Dec 09, 2022
A simple command line tool for text to image generation, using OpenAI's CLIP and a BigGAN

artificial intelligence cosmic love and attention fire in the sky a pyramid made of ice a lonely house in the woods marriage in the mountains lantern

Phil Wang 2.3k Jan 01, 2023
A Streamlit web app that generates Rick and Morty stories using GPT2.

Rick and Morty Story Generator This project uses a pre-trained GPT2 model, which was fine-tuned on Rick and Morty transcripts, to generate new stories

₸ornike 33 Oct 13, 2022
GVT is a generic translation tool for parts of text on the PC screen with Text to Speak functionality.

GVT is a generic translation tool for parts of text on the PC screen with Text to Speech functionality. I wanted to create it because the existing tools that I experimented with did not satisfy me in

Nuked 1 Aug 21, 2022
Code to use Augmented Shapiro Wilks Stopping, as well as code for the paper "Statistically Signifigant Stopping of Neural Network Training"

This codebase is being actively maintained, please create and issue if you have issues using it Basics All data files are included under losses and ea

Justin Terry 32 Nov 09, 2021
A multi-voice TTS system trained with an emphasis on quality

TorToiSe Tortoise is a text-to-speech program built with the following priorities: Strong multi-voice capabilities. Highly realistic prosody and inton

James Betker 2.1k Jan 01, 2023
Transformers-regression - Regression Bugs Are In Your Model! Measuring, Reducing and Analyzing Regressions In NLP Model Updates

Regression Free Model Update Code for the paper: Regression Bugs Are In Your Mod

Yuqing Xie 2 Feb 17, 2022
NeuralQA: A Usable Library for Question Answering on Large Datasets with BERT

NeuralQA: A Usable Library for (Extractive) Question Answering on Large Datasets with BERT Still in alpha, lots of changes anticipated. View demo on n

Victor Dibia 220 Dec 11, 2022
Weaviate demo with the text2vec-openai module

Weaviate demo with the text2vec-openai module This repository contains an example of how to use the Weaviate text2vec-openai module. When using this d

SeMI Technologies 11 Nov 11, 2022
Repositório do trabalho de introdução a NLP

Trabalho da disciplina de BI NLP Repositório do trabalho da disciplina Introdução a Processamento de Linguagem Natural da pós BI-Master da PUC-RIO. Eq

Leonardo Lins 1 Jan 18, 2022
My implementation of Safaricom Machine Learning Codility test. The code has bugs, logical I guess I made errors and any correction will be appreciated.

Safaricom_Codility Machine Learning 2022 The test entails two questions. Question 1 was on Machine Learning. Question 2 was on SQL I ran out of time.

Lawrence M. 1 Mar 03, 2022
BERT has a Mouth, and It Must Speak: BERT as a Markov Random Field Language Model

BERT has a Mouth, and It Must Speak: BERT as a Markov Random Field Language Model

303 Dec 17, 2022
The (extremely) naive sentiment classification function based on NBSVM trained on wisesight_sentiment

thai_sentiment The naive sentiment classification function based on NBSVM trained on wisesight_sentiment วิธีติดตั้ง pip install thai_sentiment==0.1.3

Charin 7 Dec 08, 2022
A music comments dataset, containing 39,051 comments for 27,384 songs.

Music Comments Dataset A music comments dataset, containing 39,051 comments for 27,384 songs. For academic research use only. Introduction This datase

Zhang Yixiao 2 Jan 10, 2022
An open source framework for seq2seq models in PyTorch.

pytorch-seq2seq Documentation This is a framework for sequence-to-sequence (seq2seq) models implemented in PyTorch. The framework has modularized and

International Business Machines 1.4k Jan 02, 2023
texlive expressions for documents

tex2nix Generate Texlive environment containing all dependencies for your document rather than downloading gigabytes of texlive packages. Installation

Jörg Thalheim 70 Dec 26, 2022