deepke/README_ENGLISH.md at 2bdd14b5f910dd14e598247855d81d72107b031a

10 KiB

Raw Blame History

A Deep Learning Based Knowledge Extraction Toolkit for Knowledge Base Population

DeepKE is a knowledge extraction toolkit supporting low-resource and document-level scenarios. It provides three functions based PyTorch, including Named Entity Recognition, Relation Extraciton and Attribute Extraction.

Online Demo

demo

Prediction

There is a demonstration of prediction.

Model Framework

Figure 1: The framework of DeepKE

DeepKE contains three modules for named entity recognition, relation extraction and attribute extraction, the three tasks respectively.
Each module has its own submodules. For example, there are standard, document-level and few-shot submodules in the attribute extraction modular.
Each submodule compose of three parts: a collection of tools, which can function as tokenizer, dataloader, preprocessor and the like, a encoder and a part for training and prediction

Quickstart

Take the fully supervised attribute extraction for example.

Download basic codes git clone https://github.com/zjunlp/DeepKE.git
Create a virtual environment (recommend anaconda) conda create -n deepke python=3.8
Enter the environment conda activate deepke
Install dependent packages
- If use deepke directly: pip install deepke
- If modify source codes before usage: run python setup.py install firstly, after modification, run python setup.py develop
Enter the corresponding directory cd DeepKE/example/re/standard
Train python run.py (Parameters for training can be changed in the conf folder)
Predict python predict.py(Parameters for prediction can be changed in the conf folder)

Requirements

python == 3.8

torch == 1.5
hydra-core == 1.0.6
tensorboard == 2.4.1
matplotlib == 3.4.1
transformers == 3.4.0
jieba == 0.42.1
scikit-learn == 0.24.1
pytorch-transformers == 1.2.0
seqeval == 1.2.2
tqdm == 4.60.0
opt-einsum==3.3.0
ujson

Introduction of Three Functions

1. Named Entity Recognition

Named entity recognition seeks to locate and classify named entities mentioned in unstructured text into pre-defined categories such as person names, organizations, locations, organizations, etc.

The data is stored in .txt files. Some instances as following:

Sentence	Person	Location	Organization
本报北京9月4日讯记者杨涌报道：部分省区人民日报宣传发行工作座谈会9月3日在4日在京举行。	杨涌	北京	人民日报
《红楼梦》是中央电视台和中国电视剧制作中心根据中国古典文学名著《红楼梦》摄制于1987年的一部古装连续剧，由王扶林导演，周汝昌、王蒙、周岭等多位红学家参与制作。	王扶林，周汝昌，王蒙，周岭	中国	中央电视台，中国电视剧制作中心
秦始皇兵马俑位于陕西省西安市，1961年被国务院公布为第一批全国重点文物保护单位，是世界八大奇迹之一。	秦始皇	陕西省，西安市	国务院

Read the detailed process in specific README
- STANDARD (Fully Supervised)
  - The standard module is implemented by the pretrained model BERT.
  - Enter DeepKE/example/ner/standard.
  - The dataset and parameters can be customized in the data folder and conf folder respectively.
  - Train: python run.py
  - Predict: python predict.py
- FEW-SHOT
  - This module is in the low-resouce scenario.
  - Enter DeepKE/example/ner/few-shot.
  - The directory where the model is loaded and saved and the configuration parameters can be cusomized in the conf folder.
  - Train with CoNLL-2003: python run.py
  - Train in the few-shot scenario: python run.py +train=few_shot. Users can modify load_path in conf/train/few_shot.yaml with the use of existing loaded model.
  - Predict: add - predict to conf/config.yaml, modify loda_path as the model path and write_path as the path where the predicted results are saved in conf/predict.yaml, and then run python predict.py

2. Relation Extraction

Relationship extraction is the task of extracting semantic relations between entities from a unstructured text.

The data is stored in .csv files. Some instances as following:

Sentence	Relation	Head	Head_offset	Tail	Tail_offset
《岳父也是爹》是王军执导的电视剧，由马恩然、范明主演。	导演	岳父也是爹	1	王军	8
《九玄珠》是在纵横中文网连载的一部小说，作者是龙马。	连载网站	九玄珠	1	纵横中文网	7
提起杭州的美景，西湖总是第一个映入脑海的词语。	所在城市	西湖	8	杭州	2

Read the detailed process in specific README
- STANDARD (Fully Supervised)
  - The standard module is implemented by common deep learning models, including CNN, RNN, Capsule, GCN, Transformer and the pretrained model.
  - Enter the DeepKE/example/re/standard folder.
  - The dataset and parameters can be customized in the data folder and conf folder respectively.
  - Train: python run.py
  - Predict: python predict.py
- FEW-SHOT
  - This module is in the low-resouce scenario.
  - Enter DeepKE/example/re/few-shot .
  - Train: python run.py Start with the model trained last time: modify train_from_saved_model in conf/train.yamlas the path where the model trained last time was saved. And the path saving logs generated in training can be customized by log_dir.
  - Predict: python predict.py
- DOCUMENT
  - Download the model train_distant.json from Google Drive to data/.
  - Enter DeepKE/example/re/document .
  - Train: python run.py Start with the model trained last time: modify train_from_saved_model in conf/train.yamlas the path where the model trained last time was saved. And the path saving logs generated in training can be customized by log_dir.
  - Predict: python predict.py

3. Attribute Extraction

Attribute extraction is to extract attributes for entities in a unstructed text.