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
Keyword-BERT: Keyword-Attentive Deep Semantic Matching

project discription An implementation of the Keyword-BERT model mentioned in my paper Keyword-Attentive Deep Semantic Matching (Plz cite this github r

1 Nov 14, 2021
Code for EMNLP2021 paper "Allocating Large Vocabulary Capacity for Cross-lingual Language Model Pre-training"

VoCapXLM Code for EMNLP2021 paper Allocating Large Vocabulary Capacity for Cross-lingual Language Model Pre-training Environment DockerFile: dancingso

Bo Zheng 15 Jul 28, 2022
Codebase for Time-series Generative Adversarial Networks (TimeGAN)

Codebase for Time-series Generative Adversarial Networks (TimeGAN)

Jinsung Yoon 532 Dec 31, 2022
Face and other object detection using OpenCV and ML Yolo

Object-and-Face-Detection-Using-Yolo- Opencv and YOLO object and face detection is implemented. You only look once (YOLO) is a state-of-the-art, real-

Happy N. Monday 3 Feb 15, 2022
Code for Neurips2021 Paper "Topology-Imbalance Learning for Semi-Supervised Node Classification".

Topology-Imbalance Learning for Semi-Supervised Node Classification Introduction Code for NeurIPS 2021 paper "Topology-Imbalance Learning for Semi-Sup

Victor Chen 40 Nov 23, 2022
AugLy is a data augmentations library that currently supports four modalities (audio, image, text & video) and over 100 augmentations

AugLy is a data augmentations library that currently supports four modalities (audio, image, text & video) and over 100 augmentations. Each modality’s augmentations are contained within its own sub-l

Facebook Research 4.6k Jan 09, 2023
(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
Kalman Filter book using Jupyter Notebook. Focuses on building intuition and experience, not formal proofs. Includes Kalman filters,extended Kalman filters, unscented Kalman filters, particle filters, and more. All exercises include solutions.

Kalman and Bayesian Filters in Python Introductory text for Kalman and Bayesian filters. All code is written in Python, and the book itself is written

Roger Labbe 13k Dec 29, 2022
A Deep learning based streamlit web app which can tell with which bollywood celebrity your face resembles.

Project Name: Which Bollywood Celebrity You look like A Deep learning based streamlit web app which can tell with which bollywood celebrity your face

BAPPY AHMED 20 Dec 28, 2021
A computer vision pipeline to identify the "icons" in Christian paintings

Christian-Iconography A computer vision pipeline to identify the "icons" in Christian paintings. A bit about iconography. Iconography is related to id

Rishab Mudliar 3 Jul 30, 2022
Learning Lightweight Low-Light Enhancement Network using Pseudo Well-Exposed Images

Learning Lightweight Low-Light Enhancement Network using Pseudo Well-Exposed Images This repository contains the implementation of the following paper

Seonggwan Ko 9 Jul 30, 2022
Dynamic vae - Dynamic VAE algorithm is used for anomaly detection of battery data

Dynamic VAE frame Automatic feature extraction can be achieved by probability di

10 Oct 07, 2022
Audio Domain Adaptation for Acoustic Scene Classification using Disentanglement Learning

Audio Domain Adaptation for Acoustic Scene Classification using Disentanglement Learning Reference Abeßer, J. & Müller, M. Towards Audio Domain Adapt

Jakob Abeßer 2 Jul 06, 2022
Official code for "InfoGraph: Unsupervised and Semi-supervised Graph-Level Representation Learning via Mutual Information Maximization" (ICLR 2020, spotlight)

InfoGraph: Unsupervised and Semi-supervised Graph-Level Representation Learning via Mutual Information Maximization Authors: Fan-yun Sun, Jordan Hoffm

Fan-Yun Sun 232 Dec 28, 2022
Send text to girlfriend in the morning

Girlfriend Text Send text to girlfriend (or really anyone with a phone number) in the morning 1. Configure your settings in utils.py. phone_number = "

Paras Adhikary 199 Oct 25, 2022
MediaPipe is a an open-source framework from Google for building multimodal

MediaPipe is a an open-source framework from Google for building multimodal (eg. video, audio, any time series data), cross platform (i.e Android, iOS, web, edge devices) applied ML pipelines. It is

Bhavishya Pandit 3 Sep 30, 2022
Bayesian Generative Adversarial Networks in Tensorflow

Bayesian Generative Adversarial Networks in Tensorflow This repository contains the Tensorflow implementation of the Bayesian GAN by Yunus Saatchi and

Andrew Gordon Wilson 1k Nov 29, 2022
Magisk module to enable hidden features on Android 12 Developer Preview 1.

Android 12 Extensions This is a Magisk module that enables hidden features on Android 12 Developer Preview 1. Features Scrolling screenshots Wallpaper

Danny Lin 384 Jan 06, 2023
Towards the D-Optimal Online Experiment Design for Recommender Selection (KDD 2021)

Towards the D-Optimal Online Experiment Design for Recommender Selection (KDD 2021) Contact 0 Jan 11, 2022

Driller: augmenting AFL with symbolic execution!

Driller Driller is an implementation of the driller paper. This implementation was built on top of AFL with angr being used as a symbolic tracer. Dril

Shellphish 791 Jan 06, 2023