OntoTune

• This is the code of the paper OntoTune: Ontology-Driven Self-training for Aligning Large Language Models (WWW2025).

In this work, we propose an ontology-driven self-training framework called OntoTune, which aims to align LLMs with ontology through in-context learning, enabling the generation of responses guided by the ontology.

🔔 News

📄arXiv • 🤗 Huggingface

• 2025-01 OntoTune is accepted by WWW 2025 !
• 2025-02 Our paper is released on arxiv !
• 2025-06 Our model is released on huggingface !

🚀 How to start

git clone https://github.com/zjukg/OntoTune.git

The code of fine-tuning is constructed based on open-sourced repo LLaMA-Factory.

Dependencies

cd LLaMA-Factory
pip install -e ".[torch,metrics]"

Data Preparation

1. The supervised instruction-tuned data generated by LLaMA3 8B for the LLM itself is placed in the link.
2. Put the downloaded OntoTune_sft.json file under LLaMA-Factory/data/ directory.
3. Evaluation datasets for hypernym discovery and medical question answering are in LLaMA-Factory/data/evaluation_HD and LLaMA-Factory/data/evaluation_QA, respectively.

Finetune LLaMA3

You need to add model_name_or_path parameter to yaml file。

cd LLaMA-Factory
llamafactory-cli train script/OntoTune_sft.yaml

🤝 Cite:

Please consider citing this paper if you find our work useful.

@inproceedings{liu2025ontotune,
  title={Ontotune: Ontology-driven self-training for aligning large language models},
  author={Liu, Zhiqiang and Gan, Chengtao and Wang, Junjie and Zhang, Yichi and Bo, Zhongpu and Sun, Mengshu and Chen, Huajun and Zhang, Wen},
  booktitle={Proceedings of the ACM on Web Conference 2025},
  pages={119--133},
  year={2025}
}