English | 中文
NeuralKG is a python-based library for diverse representation learning of knowledge graphs implementing Conventional KGEs, GNN-based KGEs, and Rule-based KGEs. We provide comprehensive documents for beginners and an online website to organize an open and shared KG representation learning community.
- Table of Contents
- 😃What's New
- Overview
- Demo
- Implemented KGEs
- Quick Start
- Reproduced Results
- Notebook Guide
- Detailed Documentation
- Citation
- NeuralKG Core Team
- [Apr, 2023] We propose a challenge about inductive KGRL at CCKS2023. Baseline is built based on NeuralKG-ind. Welcome to involve!
- [Apr, 2023] We release NeuralKG-ind for inductive Knowledge Graph Representation Learning(KGRL). The paper of NeuralKG-ind is accepted by SIGIR2023 in Demonstration Track.
- [Jan, 2023] We add the SEGNN model for our library
- [Oct, 2022] We add the DualE model for our library
- [Sep, 2022] We add the PairRE model for our library
- [Jun, 2022] We add the HAKE model for our library
- [Mar, 2022] We have provided Google Colab Tutotials help users use our library
- [Mar, 2022] We have provided a new blog about how to use NeuralKG on custom datasets
- [Feb, 2022] We have released a paper NeuralKG: An Open Source Library for Diverse Representation Learning of Knowledge Graphs, which is accepted by SIGIR2022 in Demonstration Track.
NeuralKG is built on PyTorch Lightning. It provides a general workflow of diverse representation learning on KGs and is highly modularized, supporting three series of KGEs. It has the following features:
-
Support diverse types of methods. NeuralKG, as a library for diverse representation learning of KGs, provides implementations of three series of KGE methods, including Conventional KGEs, GNN-based KGEs, and Rule-based KGEs.
-
Support easy customization. NeuralKG contains fine-grained decoupled modules that are commonly used in different KGEs, including KG Data Preprocessing, Sampler for negative sampling, Monitor for hyperparameter tuning, Trainer covering the training, and model validation.
-
long-term technical maintenance. The core team of NeuralKG will offer long-term technical maintenance. Other developers are welcome to pull requests.
There is a demonstration of NeuralKG.
| Components | Models |
|---|---|
| KGEModel | TransE, TransH, TransR, ComplEx, DistMult, RotatE, ConvE, BoxE, CrossE, SimplE, HAKE, PairRE, DualE |
| GNNModel | RGCN, KBAT, CompGCN, XTransE, SEGNN |
| RuleModel | ComplEx-NNE+AER, RUGE, IterE |
Step1 Create a virtual environment using Anaconda and enter it
conda create -n neuralkg python=3.8
conda activate neuralkgStep2 Install the appropriate PyTorch and DGL according to your cuda version
Here we give a sample installation based on cuda == 11.1
- Install PyTorch
pip install torch==1.9.1+cu111 -f https://download.pytorch.org/whl/torch_stable.html
- Install DGL
pip install dgl-cu111 dglgo -f https://data.dgl.ai/wheels/repo.html
Step3 Install package
- From Pypi
pip install neuralkg- From Source
git clone https://github.com/zjukg/NeuralKG.git
cd NeuralKG
python setup.py install# Use bash script
sh ./scripts/your-sh
# Use config
python main.py --load_config --config_path <your-config>
python main.py --test_only --checkpoint_dir <your-model-path>
NeuralKG utilizes Weights&Biases supporting various forms of hyperparameter optimization such as grid search, Random search, and Bayesian optimization. The search type and search space are specified in the configuration file in the format "*.yaml" to perform hyperparameter optimization.
The following config file displays hyperparameter optimization of the TransE on the FB15K-237 dataset using bayes search:
command:
- ${env}
- ${interpreter}
- ${program}
- ${args}
program: main.py
method: bayes
metric:
goal: maximize
name: Eval|hits@10
parameters:
dataset_name:
value: FB15K237
model_name:
value: TransE
loss_name:
values: [Adv_Loss, Margin_Loss]
train_sampler_class:
values: [UniSampler, BernSampler]
emb_dim:
values: [400, 600]
lr:
values: [1e-4, 5e-5, 1e-6]
train_bs:
values: [1024, 512]
num_neg:
values: [128, 256]
There are some reproduced model results on FB15K-237 dataset using NeuralKG as below. See more results in here
| Method | MRR | Hit@1 | Hit@3 | Hit@10 |
|---|---|---|---|---|
| TransE | 0.32 | 0.23 | 0.36 | 0.51 |
| TransR | 0.23 | 0.16 | 0.26 | 0.38 |
| TransH | 0.31 | 0.2 | 0.34 | 0.50 |
| DistMult | 0.30 | 0.22 | 0.33 | 0.48 |
| ComplEx | 0.25 | 0.17 | 0.27 | 0.40 |
| SimplE | 0.16 | 0.09 | 0.17 | 0.29 |
| ConvE | 0.32 | 0.23 | 0.35 | 0.50 |
| RotatE | 0.33 | 0.23 | 0.37 | 0.53 |
| BoxE | 0.32 | 0.22 | 0.36 | 0.52 |
| HAKE | 0.34 | 0.24 | 0.38 | 0.54 |
| PairRE | 0.35 | 0.25 | 0.38 | 0.54 |
| DualE | 0.33 | 0.24 | 0.36 | 0.52 |
| XTransE | 0.29 | 0.19 | 0.31 | 0.45 |
| RGCN | 0.25 | 0.16 | 0.27 | 0.43 |
| KBAT* | 0.28 | 0.18 | 0.31 | 0.46 |
| CompGCN | 0.34 | 0.25 | 0.38 | 0.52 |
| SEGNN | 0.36 | 0.27 | 0.39 | 0.54 |
| IterE | 0.26 | 0.19 | 0.29 | 0.41 |
*:There is a label leakage error in KBAT, so the corrected result is poor compared with the paper result. Details in deepakn97/relationPrediction#28
😃We use colab to provide some notebooks to help users use our library.
https://zjukg.github.io/NeuralKG/neuralkg.html
Please cite our paper if you use NeuralKG in your work
@inproceedings{neuralkg,
author = {Wen Zhang and
Xiangnan Chen and
Zhen Yao and
Mingyang Chen and
Yushan Zhu and
Hongtao Yu and
Yufeng Huang and
Yajing Xu and
Ningyu Zhang and
Zezhong Xu and
Zonggang Yuan and
Feiyu Xiong and
Huajun Chen},
title = {NeuralKG: An Open Source Library for Diverse Representation Learning
of Knowledge Graphs},
booktitle = {{SIGIR}},
pages = {3323--3328},
publisher = {{ACM}},
year = {2022}
}
Wen Zhang, Xiangnan Chen, Zhen Yao, Mingyang Chen, Yushan Zhu, Hongtao Yu, Yufeng Huang, Zezhong Xu, Yajing Xu, Peng Ye, Yichi Zhang, Ningyu Zhang, Guozhou Zheng, Haofen Wang, Huajun Chen