DARK

This is the official code repository for Unifying Deductive and Abductive Reasoning in Knowledge Graphs with Masked Diffusion Model (WWW 2026).

Environment

conda create -n dark python=3.9
conda activate dark
pip install -r requirements.txt 

Training

You can run the code in the following steps:

1. Sampling
2. Supervised training
3. Reinforcement learning

Step 1: Sampling

bash scripts/sample/sample_full.sh

Step 2: Supervised training

For the first-stage pretraining, set --training_mode unify.

For the second-stage supervised training for a single reasoning type, set --training_mode sft.

bash scripts/train/db.sh

or training with multi-gpu:

bash scripts/train/db-multi.sh

Step 3: Reinforcement learning

Example scripts:

bash scripts/optim/db.sh

Citation

Welcome to cite our work!

@article{gao2025unifying, title={Unifying Deductive and Abductive Reasoning in Knowledge Graphs with Masked Diffusion Model}, author={Gao, Yisen and Bai, Jiaxin and Huang, Yi and Fu, Xingcheng and Sun, Qingyun and Song, Yangqiu}, journal={arXiv preprint arXiv:2510.11462}, year={2025} }