DyMRL: Dynamic Multispace Representation Learning for Multimodal Event Forecasting in Knowledge Graph

This is the released codes of the work in the ACM Web Conference 2026 (WWW'26):

Environment

python==3.10.9
torch==2.2.1+cu118
dgl==2.1.0+cu118
tqdm==4.66.2
numpy==1.26.4

Introduction

• src: Python scripts of the DyMRL model.
• results: Model files that replicate the reported results in our paper.
• pretrain: Auxiliary linguistic and visual modality feature matrices of multimodal temporal KGs.

Training Command

cd src
CUDA_VISIBLE_DEVICES=0 python main.py --model DyMRL --dataset ICE14-IMG-TXT --bias learn --s-delta-ind --n-head 2 --rank 20 --history-len 6

cd src
CUDA_VISIBLE_DEVICES=0 python main.py --model DyMRL --dataset ICE0515-IMG-TXT --bias learn --s-delta-ind --n-head 2 --rank 20 --history-len 3

cd src
CUDA_VISIBLE_DEVICES=0 python main.py --model DyMRL --dataset ICE18-IMG-TXT --bias learn --s-delta-ind --n-head 2 --rank 20 --history-len 3

cd src
CUDA_VISIBLE_DEVICES=0 python main.py --model DyMRL --dataset GDELT-IMG-TXT --bias learn --s-delta-ind --n-head 2 --rank 20 --history-len 5

Testing Command

CUDA_VISIBLE_DEVICES=0 python main.py --model DyMRL --dataset ICE14-IMG-TXT --bias learn --s-delta-ind --n-head 2 --rank 20 --history-len 6 --test

CUDA_VISIBLE_DEVICES=0 python main.py --model DyMRL --dataset ICE0515-IMG-TXT --bias learn --s-delta-ind --n-head 2 --rank 20 --history-len 3 --test

CUDA_VISIBLE_DEVICES=0 python main.py --model DyMRL --dataset ICE18-IMG-TXT --bias learn --s-delta-ind --n-head 2 --rank 20 --history-len 3 --test

CUDA_VISIBLE_DEVICES=0 python main.py --model DyMRL --dataset GDELT-IMG-TXT --bias learn --s-delta-ind --n-head 2 --rank 20 --history-len 5 --test

Contacts

Contact us with the following email address: FrankLuis@hust.edu.cn.

Acknowledgements

The source codes take ReTIN as the backbone to implement our proposed method. Please cite both our work and ReTIN if you find this repository is helpful for your research.