CATS: Context-aware Inductive Knowledge Graph Completion with Latent Type Constraints and Subgraph Reasoning

This repository provides the official implementation of the paper "Context-aware Inductive Knowledge Graph Completion with Latent Type Constraints and Subgraph Reasoning" (To appear in AAAI2025).

Experiment Environment Setup

Create a python environment and install the required packages. We suggest you use Python 3.10 with PyTorch 2.2.2. For detailed Python package versions, you may refer to the suggested settings listed in requirements.txt from LLaMA-Factory. (VLLM is not required)

pip install -r requirements.txt

Additionally, install sentence_transformers:

pip install sentence_transformers

Dataset

1. Download the full dataset and LLM instructions from the following link:

• Dataset & Instructions

2. Copy the two subfolders "datasets" and "instructions" into the project directory.

Alternatively, you can construct the LLM instruction prompts by executing python build_instruction.py.

LLM Setup

You may download LLM checkpoints from the following links: Our experimental results can be reproduced with the Qwen2-7B-Instruct LLM.

• Qwen2-7B-Instruct
• Meta-Llama-3-8B-Instruct

Please update the default value of LLM_PATH in script data_manager.py with your local model path.

Intruction-tuning

Please refer to the official document of LLaMA-Factory to conduct LLM supervised fine-tuning with the provided prompts. You may need to specify the prompt path and other training settings in a configuration file. Detailed hyper-parameters are provided in our paper.

Inference

The following command evaluates the model performance. You may alter the parameters below to test the model in different (transductive, inductive, and few-shot) scenarios.

python3 prediction.py --dataset FB15k-237-subset --setting inductive --training_size full --model_name {model_path_after_sft} --prompt_type CATS --subgraph_type together --path_type degree

Citation

If you find this code useful, please consider citing the following paper.

@misc{li2024contextawareinductiveknowledgegraph,
      title={Context-aware Inductive Knowledge Graph Completion with Latent Type Constraints and Subgraph Reasoning}, 
      author={Muzhi Li and Cehao Yang and Chengjin Xu and Zixing Song and Xuhui Jiang and Jian Guo and Ho-fung Leung and Irwin King},
      year={2024},
      eprint={2410.16803},
      archivePrefix={arXiv},
      primaryClass={cs.AI},
      url={https://arxiv.org/abs/2410.16803}, 
}