This is the code repo for Advancing Abductive Reasoning in Knowledge Graphs through Complex Logical Hypothesis Generation

• Paper: https://arxiv.org/abs/2312.15643
• Slides: Google drive

Environment

conda create -n akgr python=3.10
conda activate akgr
conda install pytorch==2.0.1 torchvision==0.15.2 torchaudio==2.0.2 pytorch-cuda=11.8 -c pytorch -c nvidia
pip install -r requirements.txt 

Training

As described in the paper, the Reinforcement Learning from Knowledge Graph Feedback (RLF-KG) pipeline comprises the following steps:

1. Sampling
2. Supervised training
3. Reinforcement learning

Step 1: Sampling

bash scripts/sample/sample_full.sh

See Example Data and Checkpoints

Step 2: Supervised training

Example scripts:

bash scripts/train/fb-t5.sh
bash scripts/train/db-t5.sh
bash scripts/train/wn-t5.sh
bash scripts/train/fb-g2.sh
bash scripts/train/db-g2.sh
bash scripts/train/wn-g2.sh

Step 3: Reinforcement learning

Example scripts:

bash scripts/optim/fb-t5-0.0.sh
bash scripts/optim/fb-t5-0.2.sh
bash scripts/optim/db-t5-0.0.sh
bash scripts/optim/db-t5-0.2.sh
bash scripts/optim/wn-t5-0.0.sh
bash scripts/optim/wn-t5-0.2.sh
bash scripts/optim/fb-g2-0.0.sh
bash scripts/optim/fb-g2-0.2.sh
bash scripts/optim/db-g2-0.0.sh
bash scripts/optim/db-g2-0.2.sh
bash scripts/optim/wn-g2-0.0.sh
bash scripts/optim/wn-g2-0.2.sh

Evaluation

Example scripts:

bash scripts/test/fb-t5.sh
bash scripts/test/db-t5.sh
bash scripts/test/wn-t5.sh
bash scripts/test/fb-g2.sh
bash scripts/test/db-g2.sh
bash scripts/test/wn-g2.sh

bash scripts/optim-test/fb-t5-0.0.sh
bash scripts/optim-test/fb-t5-0.2.sh
bash scripts/optim-test/db-t5-0.0.sh
bash scripts/optim-test/db-t5-0.2.sh
bash scripts/optim-test/wn-t5-0.0.sh
bash scripts/optim-test/wn-t5-0.2.sh
bash scripts/optim-test/fb-g2-0.0.sh
bash scripts/optim-test/fb-g2-0.2.sh
bash scripts/optim-test/db-g2-0.0.sh
bash scripts/optim-test/db-g2-0.2.sh
bash scripts/optim-test/wn-g2-0.0.sh
bash scripts/optim-test/wn-g2-0.2.sh

See Example Data and Checkpoints

Example Data and Checkpoints

Sampled data: Download Onedrive to sampled_data under the root.

Checkpoints: Download Onedrive to checkpoints under the root.

Citation

@misc{bai2024advancingabductivereasoningknowledge,
      title={Advancing Abductive Reasoning in Knowledge Graphs through Complex Logical Hypothesis Generation}, 
      author={Jiaxin Bai and Yicheng Wang and Tianshi Zheng and Yue Guo and Xin Liu and Yangqiu Song},
      year={2024},
      eprint={2312.15643},
      archivePrefix={arXiv},
      primaryClass={cs.AI},
      url={https://arxiv.org/abs/2312.15643}, 
}