PRoH: Dynamic Planning and Reasoning over Knowledge Hypergraphs for Retrieval-Augmented Generation (WWW 2026)

PRoH is a Knowledge Hypergraph–based retrieval-augmented generation (RAG) framework for multi-hop question answering. It integrates context-aware planning with structured question decomposition, enabling adaptive planning and reasoning on knowledge hypergraphs with beyond binary relational structures.

---

Directory Layout

.
├── PRoH/                 # Main source code
├── eval/                 # Evaluation scripts and metrics
├── contexts/             # Source files for Knowledge Hypergraphs
├── questions/            # Provided QA datasets
├── construct.py          # Build a Knowledge Hypergraph from raw text
├── query.py              # Execute a single query
├── script_query.py       # Batch querying
├── test_construct.sh     # Example script for KH construction
├── test_query.sh         # Example pipeline for query + evaluation
├── openai_api_key.txt    # OpenAI API key file (user-provided)
├── requirements.txt      # Python dependencies
└── Readme.md             # This file

---

Environment Setup

conda create -n proh python=3.11
conda activate proh
pip install -r requirements.txt

Quick Start

Knowledge HyperGraph Construction

Construct a Knowledge Hypergraph (KH) from source text. Five supported domains: agriculture, cs, hypertension, legal, mix.

bash test_construct.sh example

Knowledge HyperGraph Query

Run a single query over the constructed KH

python query.py >> query.out 2>&1

Knowledge HyperGraph Query & Evaluation (pipline)

Query the Knowledge HyperGraph and evaluate the answer as a pipline

bash test_query.sh example

How to cite

If you are interested or inspired by this work, you can cite us by:

@article{zai2025proh,
  title={PRoH: Dynamic Planning and Reasoning over Knowledge Hypergraphs for Retrieval-Augmented Generation},
  author={Zai, Xiangjun and Tan, Xingyu and Wang, Xiaoyang and Liu, Qing and Xu, Xiwei and Zhang, Wenjie},
  journal={arXiv preprint arXiv:2510.12434},
  year={2025}
}

Acknowledgement

This repo benefits from HyperGraphRAG. Thanks for their wonderful works.