RManLuo / NP-FKGC

Official code implementation for SIGIR 23 paper Normalizing Flow-based Neural Process for Few-Shot Knowledge Graph Completion

NP-FKGC

Official code implementation for SIGIR 23 paper Normalizing Flow-based Neural Process for Few-Shot Knowledge Graph Completion

Requirement

pytorch==1.11
tqdm==4.64
normflows==1.4
dgl==0.9.0
tensorboardx==2.5.1

Note: Please make sure dgl==0.9.0 and use CUDA, our codes rely on a small bug of dgl for running.

Environment

• python 3.8
• Ubuntu 22.04
• RTX3090/A100
• Memory 32G/128G

Dataset & Checkpoint

Original Dataset

• NELL/WIKI
• FB15K-237

Processed Dataset

• Dataset
• Checkpoint

Download the datasets and extract to the project root folder.

Train

NELL (3090)

python main.py --dataset NELL-One --data_path ./NELL --few 5 --data_form Pre-Train --prefix np_rgcn_attn_planar_nellone_5shot_intrain --device 0 --batch_size 128 --flow Planar --g_batch 1024

WIKI (A100)

python main.py --dataset Wiki-One --data_path ./Wiki --few 5 --data_form Pre-Train --prefix np_rgcn_attn_planar_wiki_5shot_intrain_g_batch_1024_eval_8 --device 0 --batch_size 64 --flow Planar -dim 50 --g_batch 1024 --eval_batch 8 --eval_epoch 4000

FB15K-237 (3090)

python main.py --dataset FB15K-One --data_path ./FB15K --few 5 --data_form Pre-Train --prefix np_rgcn_attn_planar_fb15k_5shot_intrain --device 0 --batch_size 128 --flow Planar --g_batch 1024 --eval_batch_size 128 --K 14

Eval

Download the checkpoint and extract to the state/ folder.

NELL

python main.py --dataset NELL-One --data_path ./NELL --few 5 --data_form Pre-Train --prefix np_rgcn_attn_planar_nellone_5shot_intrain_0.46 --device 0 --batch_size 128 --flow Planar --g_batch 1024 --step test

WIKI

python main.py --dataset Wiki-One --data_path ./Wiki --few 5 --data_form Pre-Train --prefix np_rgcn_attn_planar_wiki_5shot_intrain_g_batch_1024_eval_8_0.503 --device 0 --batch_size 64 --flow Planar -dim 50 --g_batch 1024 --eval_batch 8 --eval_epoch 4000 --step test

FB15K-237

python main.py --dataset FB15K-One --data_path ./FB15K --few 5 --data_form Pre-Train --prefix np_rgcn_attn_planar_fb15k_5shot_intrain_0.536 --device 0 --batch_size 128 --flow Planar --g_batch 1024 --eval_batch_size 128 --K 14 --step test

Results

5-shot FKGC results

Dataset	MRR	Hits@10	Hits@5	Hits@1
NELL	0.460	0.494	0.471	0.437
WIKI	0.503	0.668	0.599	0.423
FB15K-237	0.538	0.671	0.593	0.476

See full results in our paper.

Citations

If you use this repo, please cite the following paper.

@inproceedings{
 luo2023npfkgc,
 title={Normalizing Flow-based Neural Process for Few-Shot Knowledge Graph Completion},
 author={Linhao Luo, Yuan-Fang Li, Gholamreza Haffari, and Shirui Pan},
 booktitle={The 46th International ACM SIGIR Conference on Research and Development in Information Retrieval},
 year={2023}
}

Acknowledgement

This repo is mainly based on GANA. We thank the authors for their great works.