Relation-enhanced Negative Sampling for Multimodal Knowledge Graph Completion

This is the code of the paper Relation-enhanced Negative Sampling for Multimodal Knowledge Graph Completion for ACM MM 2022.

Requirements

• pytorch == 1.10.1
• numpy == 1.20.3

Datasets

The source images and triples of MMKB-DB15K are from mmkb.

The source text of three datasets are from DBpedia.

The embeddings and raw data can be downloaded in the Google Drive

Usage

mkdir data models results

put the datasets in ./data and

python run_gumbel.py --do_train --do_valid --do_test --data_path=data/MMKB-DB15K --model=TransE -n=20 -d=200 -g=6 -a=0.5 \
        -r=0.0 -lr=0.0001 -kca_lr=0.0001 --sample_method=gumbel  --pre_sample_num=1500  --loss_rate=100 --exploration_temp=10 \
        --gpu=0  --max_steps=100000 --valid_steps=10000 -b=400

This code refers to the code of RotatE and Nscaching.

Citation

If you find this codebase useful in your research, please cite the following paper.

@inproceedings{xu2022relation,
  title={Relation-enhanced Negative Sampling for Multimodal Knowledge Graph Completion},
  author={Xu, Derong and Xu, Tong and Wu, Shiwei and Zhou, Jingbo and Chen, Enhong},
  booktitle={Proceedings of the 30th ACM International Conference on Multimedia},
  pages={3857--3866},
  year={2022}
}