Chinese named entity recognization with BiLSTM using Keras

Related tags

Deep LearningBiLSTM-NER-Keras
Overview

Chinese named entity recognization (Bilstm with Keras)

Project Structure

./
├── README.md
├── data
│   ├── README.md
│   ├── data							数据集
│   │   ├── test.txt
│   │   └── train.txt
│   ├── plain_text.txt
│   └── vocab.txt                       词表
├── evaluate
│   ├── __init__.py
│   └── f1_score.py                     计算实体F1得分
├── keras_contrib                       keras_contrib包，也可以pip装
├── log                                 训练nohup日志
│   ├── __init__.py
│   └── nohup.out
├── model                               模型
│   ├── BiLSTMCRF.py
│   ├── __init__.py
│   └── __pycache__
├── predict                             输出预测
│   ├── __init__.py
│   ├── __pycache__
│   ├── predict.py
│   └── predict_process.py
├── preprocess                          数据预处理
│   ├── README.md
│   ├── __pycache__
│   ├── convert_jsonl.py
│   ├── data_add_line.py
│   ├── generate_vocab.py               生成词表
│   ├── process_data.py                 数据处理转换
│   ├── splite.py
│   └── vocab.py                        词表对应工具
├── public
│   ├── __init__.py
│   ├── __pycache__
│   ├── config.py                       训练设置
│   ├── generate_label_id.py            生成label2id文件
│   ├── label2id.json                   标签dict
│   ├── path.py                         所有路径
│   └── utils.py                        小工具
├── report
│   └── report.out                      F1评估报告
├── train.py
└── weight                              保存的权重
    └── bilstm_ner.h5

52 directories, 214 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tab（"\t"）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. 修改public/path.py中的地址
  3. 使用public/generate_label_id.py生成label2id.txt文件，将其中的内容填到preprocess/vocab.py的get_tag2index中。注意：序号必须从0开始
  4. 修改public/config.py中的MAX_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_LEN）
  5. 运行preprocess/generate_vocab.py生成词表，词表按词频生成
  6. 根据需要修改BiLSTMCRF.py模型结构
  7. 修改public/config.py的参数
  8. 训练前debug看下train_data,train_label对不对
  9. 训练

Model

_________________________________________________________________
Layer (type)                 Output Shape              Param #
=================================================================
input_1 (InputLayer)         (None, None)              0
_________________________________________________________________
embedding_1 (Embedding)      (None, None, 128)         81408
_________________________________________________________________
bidirectional_1 (Bidirection (None, None, 256)         263168
_________________________________________________________________
dropout_1 (Dropout)          (None, None, 256)         0
_________________________________________________________________
bidirectional_2 (Bidirection (None, None, 128)         164352
_________________________________________________________________
dropout_2 (Dropout)          (None, None, 128)         0
_________________________________________________________________
time_distributed_1 (TimeDist (None, None, 29)          3741
_________________________________________________________________
dropout_3 (Dropout)          (None, None, 29)          0
_________________________________________________________________
crf_1 (CRF)                  (None, None, 29)          1769
=================================================================
Total params: 514,438
Trainable params: 514,438
Non-trainable params: 0
_________________________________________________________________

Train

运行train.py

Epoch 1/500
806/806 [==============================] - 15s 18ms/step - loss: 2.4178 - crf_viterbi_accuracy: 0.9106

Epoch 00001: loss improved from inf to 2.41777, saving model to /home/bureaux/Projects/BiLSTMCRF_TimeDistribute/weight/bilstm_ner.h5
Epoch 2/500
806/806 [==============================] - 10s 13ms/step - loss: 0.6370 - crf_viterbi_accuracy: 0.9106

Epoch 00002: loss improved from 2.41777 to 0.63703, saving model to /home/bureaux/Projects/BiLSTMCRF_TimeDistribute/weight/bilstm_ner.h5
Epoch 3/500
806/806 [==============================] - 11s 14ms/step - loss: 0.5295 - crf_viterbi_accuracy: 0.9106

Epoch 00003: loss improved from 0.63703 to 0.52950, saving model to /home/bureaux/Projects/BiLSTMCRF_TimeDistribute/weight/bilstm_ner.h5
Epoch 4/500
806/806 [==============================] - 11s 13ms/step - loss: 0.4184 - crf_viterbi_accuracy: 0.9064

Epoch 00004: loss improved from 0.52950 to 0.41838, saving model to /home/bureaux/Projects/BiLSTMCRF_TimeDistribute/weight/bilstm_ner.h5
Epoch 5/500
806/806 [==============================] - 12s 14ms/step - loss: 0.3422 - crf_viterbi_accuracy: 0.9104

Epoch 00005: loss improved from 0.41838 to 0.34217, saving model to /home/bureaux/Projects/BiLSTMCRF_TimeDistribute/weight/bilstm_ner.h5
Epoch 6/500
806/806 [==============================] - 10s 13ms/step - loss: 0.3164 - crf_viterbi_accuracy: 0.9106

Epoch 00006: loss improved from 0.34217 to 0.31637, saving model to /home/bureaux/Projects/BiLSTMCRF_TimeDistribute/weight/bilstm_ner.h5
Epoch 7/500
806/806 [==============================] - 10s 12ms/step - loss: 0.3003 - crf_viterbi_accuracy: 0.9111

Epoch 00007: loss improved from 0.31637 to 0.30032, saving model to /home/bureaux/Projects/BiLSTMCRF_TimeDistribute/weight/bilstm_ner.h5
Epoch 8/500
806/806 [==============================] - 10s 12ms/step - loss: 0.2906 - crf_viterbi_accuracy: 0.9117

Epoch 00008: loss improved from 0.30032 to 0.29058, saving model to /home/bureaux/Projects/BiLSTMCRF_TimeDistribute/weight/bilstm_ner.h5
Epoch 9/500
806/806 [==============================] - 9s 12ms/step - loss: 0.2837 - crf_viterbi_accuracy: 0.9118

Epoch 00009: loss improved from 0.29058 to 0.28366, saving model to /home/bureaux/Projects/BiLSTMCRF_TimeDistribute/weight/bilstm_ner.h5
Epoch 10/500
806/806 [==============================] - 9s 11ms/step - loss: 0.2770 - crf_viterbi_accuracy: 0.9142

Epoch 00010: loss improved from 0.28366 to 0.27696, saving model to /home/bureaux/Projects/BiLSTMCRF_TimeDistribute/weight/bilstm_ner.h5
Epoch 11/500
806/806 [==============================] - 10s 12ms/step - loss: 0.2713 - crf_viterbi_accuracy: 0.9160

Evaluate

运行evaluate/f1_score.py

100%|█████████████████████████████████████████| 118/118 [00:38<00:00,  3.06it/s]
TP: 441
TP+FP: 621
precision: 0.7101449275362319
TP+FN: 604
recall: 0.7301324503311258
f1: 0.72

classification report:
              precision    recall  f1-score   support

     ANATOMY       0.74      0.75      0.74       220
    BOUNDARY       1.00      0.75      0.86         8
     DENSITY       0.78      0.88      0.82         8
    DIAMETER       0.82      0.88      0.85        16
     DISEASE       0.54      0.72      0.62        43
   LUNGFIELD       0.83      0.83      0.83         6
      MARGIN       0.57      0.67      0.62         6
      NATURE       0.00      0.00      0.00         6
       ORGAN       0.62      0.62      0.62        13
    QUANTITY       0.88      0.87      0.87        83
       SHAPE       1.00      0.43      0.60         7
        SIGN       0.66      0.65      0.65       189
     TEXTURE       0.75      0.43      0.55         7
   TREATMENT       0.25      0.33      0.29         9

   micro avg       0.71      0.71      0.71       621
   macro avg       0.67      0.63      0.64       621
weighted avg       0.71      0.71      0.71       621

Predict

运行predict/predict_bio.py

Owner
GitHub Repository
A Planar RGB-D SLAM which utilizes Manhattan World structure to provide optimal camera pose trajectory while also providing a sparse reconstruction containing points, lines and planes, and a dense surfel-based reconstruction.

ManhattanSLAM Authors: Raza Yunus, Yanyan Li and Federico Tombari ManhattanSLAM is a real-time SLAM library for RGB-D cameras that computes the camera

117 Dec 28, 2022
Duke Machine Learning Winter School: Computer Vision 2022

mlwscv2002 Welcome to the Duke Machine Learning Winter School: Computer Vision 2022! The MLWS-CV includes 3 hands-on training sessions on implementing

Duke + Data Science (+DS) 9 May 25, 2022
Replication Package for AequeVox:Automated Fariness Testing for Speech Recognition Systems

AequeVox Replication Package for AequeVox:Automated Fariness Testing for Speech Recognition Systems README under development. Python Packages Required

Sai Sathiesh 2 Aug 28, 2022
CT Based COVID 19 Diagnose by Image Processing and Deep Learning

This project proposed the deep learning and image processing method to undertake the diagnosis on 2D CT image and 3D CT volume.

1 Feb 08, 2022
Estimation of human density in a closed space using deep learning.

Siemens HOLLZOF challenge - Human Density Estimation Add project description here. Installing Dependencies: Install Python3 either system-wide, user-w

3 Aug 08, 2021
✨风纪委员会自动投票脚本，利用Github Action帮你进行裁决操作（为了让其他风纪委员有案件可判，本程序从中午12点才开始运行，有需要请自己修改运行时间）

风纪委员会自动投票 本脚本通过使用Github Action来实现B站风纪委员的自动投票功能，喜欢请给我点个STAR吧！ 如果你不是风纪委员，在符合风纪委员申请条件的情况下，本脚本会自动帮你申请 投票时间是早上八点，如果有需要请自行修改.github/workflows/Judge.yml中的时间，

Pesy Wu 25 Feb 17, 2021
[PNAS2021] The neural architecture of language: Integrative modeling converges on predictive processing

The neural architecture of language: Integrative modeling converges on predictive processing Code accompanying the paper The neural architecture of la

Martin Schrimpf 36 Dec 01, 2022
Dataset Condensation with Contrastive Signals

Dataset Condensation with Contrastive Signals This repository is the official implementation of Dataset Condensation with Contrastive Signals (DCC). T

3 May 19, 2022
Python script to download the celebA-HQ dataset from google drive

download-celebA-HQ Python script to download and create the celebA-HQ dataset. WARNING from the author. I believe this script is broken since a few mo

133 Dec 21, 2022
This repository contains the code for the paper ``Identifiable VAEs via Sparse Decoding''.

Sparse VAE This repository contains the code for the paper ``Identifiable VAEs via Sparse Decoding''. Data Sources The datasets used in this paper wer

Gemma Moran 17 Dec 12, 2022
Parameterising Simulated Annealing for the Travelling Salesman Problem

Parameterising Simulated Annealing for the Travelling Salesman Problem

Gary Sun 55 Jun 15, 2022
Relative Uncertainty Learning for Facial Expression Recognition

Relative Uncertainty Learning for Facial Expression Recognition The official implementation of the following paper at NeurIPS2021: Title: Relative Unc

35 Dec 28, 2022
Super-BPD: Super Boundary-to-Pixel Direction for Fast Image Segmentation (CVPR 2020)

Super-BPD for Fast Image Segmentation (CVPR 2020) Introduction We propose direction-based super-BPD, an alternative to superpixel, for fast generic im

189 Dec 07, 2022
Language Models for the legal domain in Spanish done @ BSC-TEMU within the "Plan de las Tecnologías del Lenguaje" (Plan-TL).

Spanish legal domain Language Model ⚖️ This repository contains the page for two main resources for the Spanish legal domain: A RoBERTa model: https:/

Plan de Tecnologías del Lenguaje - Gobierno de España 12 Nov 14, 2022
Cervix ROI Segmentation Using U-NET

Cervix ROI Segmentation Using U-NET Overview This code illustrate how to segment the ROI in cervical images using U-NET. The ROI here meant to include

Scotty Kwok 35 Sep 14, 2022
CLIPort: What and Where Pathways for Robotic Manipulation

CLIPort CLIPort: What and Where Pathways for Robotic Manipulation Mohit Shridhar, Lucas Manuelli, Dieter Fox CoRL 2021 CLIPort is an end-to-end imitat

246 Dec 11, 2022
QuakeLabeler is a Python package to create and manage your seismic training data, processes, and visualization in a single place — so you can focus on building the next big thing.

QuakeLabeler Quake Labeler was born from the need for seismologists and developers who are not AI specialists to easily, quickly, and independently bu

Hao Mai 15 Nov 04, 2022
This repository contains the source codes for the paper AtlasNet V2 - Learning Elementary Structures.

AtlasNet V2 - Learning Elementary Structures This work was build upon Thibault Groueix's AtlasNet and 3D-CODED projects. (you might want to have a loo

Théo Deprelle 123 Nov 11, 2022
This is a Python Module For Encryption, Hashing And Other stuff

EnroCrypt This is a Python Module For Encryption, Hashing And Other Basic Stuff You Need, With Secure Encryption And Strong Salted Hashing You Can Do

5 Sep 15, 2022
(IEEE TIP 2021) Regularized Densely-connected Pyramid Network for Salient Instance Segmentation

RDPNet IEEE TIP 2021: Regularized Densely-connected Pyramid Network for Salient Instance Segmentation PyTorch training and testing code are available.

Yu-Huan Wu 41 Oct 21, 2022