DMIS at BioCreative IX MedHopQA Track

This repository contains the code implementation of our methodology for generating answers to multi-step reasoning questions in the healthcare and biomedical domains, as introduced in the BioCreative IX MedhopQA Challenge.

Overview & Quick Links

• Task description: The MedHopQA task requires the generation of answers to multi-step reasoning questions in the healthcare and biomedical domains. For detailed information, please visit to the official BC9 website (link) or see the overview paper (PDF).
• System description: Please check out our paper listed below.
  • This paper provides a brief description of our system.

Requirements

This repository has been tested on Python==3.10, PyTorch==2.10, and Transformers==4.57.6. See the requirements.txt file and instructions below.

# Download this project
git clone https://github.com/dmis-lab/bc9-medhop-qa.git
cd bc9-medhop-qa

# Create a conda environment
conda create -n bc9 python=3.10
conda activate bc9

# Install all requirements
pip install -r requirements.txt

File Structure

bc9-medhop-qa
├── core
│   ├── llm
│   │   ├── gemini.py
│   │   ├── gpt_4o.py
│   │   ├── gpt_5_2.py
│   │   └── o3.py
│   ├── medhop_dataset.py
│   ├── reranker
│   │   ├── medcpt_reranker.py
│   │   ├── qwen3_reranker.py
│   │   └── reranker_base.py
│   └── retriever
│       ├── dense_wikipedia.py
│       ├── elastic_wikipedia.py
│       └── retriever_base.py
├── data
│   └── MedHopQA_Test_Dataset.csv
├── decision_making.py
├── output
│   ├── result_decision_making.csv
│   ├── result_q2d.csv
│   ├── result_rag.csv
│   ├── result_rationale.csv
│   └── result_web_search.csv
├── prompt
│   ├── decision_making.py
│   ├── query2doc.py
│   ├── rationale.py
│   └── web_search.py
├── query2doc.py
├── rationale.py
├── README.md
├── utils
│   ├── merge_csvs.py
│   └── normalize_long_answer.py
└── web_search.py

Excecution

# 1. Run Query2Doc
python query2doc.py

# 2. Run Rationale
python ratinoale.py

# 3. Run Web Search
python web_search.py

# 4. Merge outputs
python utils/merge_csvs.py

# 5. Run decision-making
python decision-making.py

• Note: The Elasticsearch server and Wikipedia embeddings required for retrieval must be configured independently.

References

Please cite the papers below if you use our code, method, or if your work is inspired by ours.

@inproceedings{jung2025dmis,
  author    = {Jung, J. and Hwang, H. and Park, Y. and Song, M. and Yoon, J. and Hwang, H. and Lee, S. and Sohn, J. and Kang, J.},
  title     = {DMIS Lab at MedHopQA-2025: Ensemble Multi-Retrieval Methodologies with Reasoning Language Model Decision},
  booktitle = {Proceedings of the BioCreative IX Challenge and Workshop (BC9): Large Language Models for Clinical and Biomedical NLP at the International Joint Conference on Artificial Intelligence (IJCAI)},
  year      = {2025},
  doi       = {10.5281/zenodo.16875789},
  url       = {https://doi.org/10.5281/zenodo.16875789}
}

Also, it should be noted that appropriate references must be cited when using the MedHopQA dataset or citing BC9 challenge results, etc.