HARIS

This repository provides the code and data for the paper:
"Coordinating Search-Informed Reasoning and Reasoning-Guided Search in Claim Verification."

Note: We are still working on the project. Please kindly check back for updates.

Data

We mainly use the HOVER and EX-FEVER datasets for experiments.

• Test data: test_data/
• Main results: exp_results/

Retrieval Corpus

The retrieval system and Wikipedia corpus setup are based on the FlashRAG framework.

Model Checkpoints

Checkpoints for the reasoning and search agents are available at:

• Reasoning Agent
• Search Agent

Hosting the Agents

1. Host the Search Agent

After setting up the retrieval system, you can host the search agent with:

export SEARCH_CORPUS_URL=http://127.0.0.1:8001/search
export CUDA_VISIBLE_DEVICES=0

python host_low.py \
    --model_path [path_to_search_agent] \
    --port 8002

Quick test:

curl -X POST http://127.0.0.1:8002/question \
     -H "Content-Type: application/json" \
     -d '{"question": "What is the capital of France?"}'

2. Host the Reasoning Agent

export SEARCH_CORPUS_URL=http://127.0.0.1:8000/search
export SEARCH_AGENT_URL=http://127.0.0.1:8002/question
export CUDA_VISIBLE_DEVICES=1

python host_high.py \
    --model_path [path_to_reasoning_agent] \
    --port 8003

Quick test:

curl -X POST http://127.0.0.1:8003/verify \
     -H "Content-Type: application/json" \
     -d '{"claim": "The movie that starred Augustus Prew was directed by the founder of animation studio Klasky Csupo."}'

Alternatively, you can run reasoning agent inference using haris_infer.py.

Citation

Feel free to cite this work if you find our work useful.

@article{hu2025coordinatingsearchinformedreasoningreasoningguided,
  title={Coordinating Search-Informed Reasoning and Reasoning-Guided Search in Claim Verification},
  author={Qisheng Hu and Quanyu Long and Wenya Wang},
  journal={arXiv preprint arXiv:2506.07528},
  year={2025},
  url={https://arxiv.org/abs/2506.07528},
}