gtfintechlab/KG-QAGen-D

KG-QAGen: A Knowledge-Graph-Based Framework for Systematic Question Generation and Long-Context LLM Evaluation

KG‑QAGen is a framework that leverages structured annotations of large documents to build knowledge graphs and systematically extract QA pairs at controlled difficulty levels, enabling fine‑grained evaluation of long‑context LLMs.

KG‑QAGen‑D Dataset

We produce KG‑QAGen‑D, a 20,139-question benchmark derived from 170 SEC credit agreements (2013–2022). Each QA pair is tagged with a composite complexity level (L = #hops + #set‑ops + plurality), split into Easy, Medium, and Hard.

Leaderboard & Evaluation Platform

To facilitate reproducibility and future research, we release the KG‑QAGen‑D dataset under a CC-BY-NC-ND 4.0 license. The dataset is divided into development and test sets as follows:

Stats	Dev	Test	Total
# Documents	40	130	170
# Questions per Doc (Min)	1	1	1
# Questions per Doc (Avg)	14.75	23.49	21.44
# Questions per Doc (Max)	83	428	428
# Easy Questions	1,499	5,051	6,550
# Medium Questions	2,680	10,203	12,883
# Hard Questions	239	467	706
Total Questions	4,418	15,721	20,139

• Development Set (~25%): 40 documents and 4,418 QA pairs are publicly released to support model development and validation.
• Test Set (~75%): 130 documents and 15,721 QA pairs are not released to prevent data contamination and ensure fair evaluation (questions are released for the leaderboard).

Online Leaderboard

We will host an evaluation leaderboard on Hugging Face upon acceptance of the paper.

Contact

For questions or issues, please reach out to:

• Nikita Tatarinov: [email protected]
• Agam Shah: [email protected]

Running the codes

1. Ensure the files from HuggingFace are placed in data/questions directory. For inference only, files without ground-truth answers are sufficient. To run benchmarking, ground-truth answers are also required.

2. Ensure you have conda or Anaconda/Miniconda installed.

3. In your terminal, navigate to the project directory.

4. To create the necessary conda environment, run:

conda env create -f environment.yml

5. Once environment creation finishes, activate it:

conda activate kgqagen

6. Customize inference/config.py file to run benchmarking in a specific setting.

   • Customize QUESTION_FILE to change the complexity level of questions to benchmark on.

   • Customize LLM_PROVIDER and MODEL_NAME to change the model to benchmark.

   • Customize TESTING_REGIME to change the benchmarking setting.

     • Use "FULL" for benchmarking LLMs with entire documents.

     • Use "RAG" for benchmarking LLMs with RAG (retrieval-augmented generation).

     • Use "GOLD" for benchmarking LLMs with pieces of documents containing the answer (Oracle setting).

7. For inference, navigate to inference directory.

   • (Optional) For benchmarking with RAG, create a vectore store:

   python vector_db_builder.py
   

   • Run the inference:

   python main.py
   

8. For benchmarking, navigate to benchmarking directory.

   • Regroup the questions into "Easy", "Medium" and "Hard" categories:

   python restructure.py
   

   • Run the metrics calculation. Remove --llm-as-a-judge to calculate all metrics but LLM-as-a-Judge. Add --llm-provider openai to use OpenAI's gpt-4o as a judge (use OpenAI API key then).

   python metrics.py --llm-as-a-judge [Gemini API key]
   

Citation

If you use KG‑QAGen in your work, please cite:

@inproceedings{kgqagen2025,
  title     = {{KG‑QAGen}: A Knowledge‑Graph‑Based Framework for Systematic Question Generation and Long‑Context LLM Evaluation},
  author    = {Tatarinov, Nikita and Kannan, Vidhyakshaya and Srinivasa, Haricharana and Raj, Arnav and Anand, Harpreet and Singh, Varun and Luthra, Aditya and Lade, Ravij and Shah, Agam and Chava, Sudheer},
  booktitle = {NeurIPS Dataset and Benchmark},
  year      = {2025},
  url       = {https://github.com/gtfintechlab/KG-QAGen}
}