This repository contains the code and resources for the RTQA framework, as described in the paper: RTQA: Recursive Thinking for Complex Temporal Knowledge Graph Question Answering with Large Language Models.
- Our paper has been accepted to EMNLP 2025 main 🎉
- Release the code and resources before 2025-09-30
- Our paper is released on arxiv !
git clone https://github.com/zjukg/RTQA.git
conda create -n RTQA python=3.9.21
conda activate RTQA
pip install -r requirements.txtThe RTQA framework uses the MultiTQ and TimelineKGQA datasets for evaluation. Below are instructions to download these datasets:
- Visit the https://github.com/czy1999/MultiTQ.
git clone https://github.com/czy1999/MultiTQ.git
cd MultiTQ/data
unzip Dataset.zip- Alternatively, download the dataset directly from Hugging Face:
🤗Datasets Link: https://huggingface.co/datasets/chenziyang/MultiTQ
Visit the https://github.com/PascalSun/TimelineKGQA/tree/main/Datasets
git clone https://github.com/PascalSun/TimelineKGQA.git
cd DatasetsNote: The TimelineKGQA dataset is generated based on ICEWS Actor and CronQuestions KG. We only use the CronQuestions KG part.
cd MultiTQ/TimelineKGQA
cd TemQuesDecom
python 0_get_prompt.py
phthon 1_query.py
python 2_combine.py
python 3_self_check.py
python 4_postprocess.py
phthon 5_postprocess_tree.py
cd ../RecursiveSolver
python 1_built_tree_time.py
python 2_run.py
python 3_get_f1.pyPlease consider citing this paper if you find our work useful.
@misc{gong2025rtqarecursivethinking,
title={RTQA : Recursive Thinking for Complex Temporal Knowledge Graph Question Answering with Large Language Models},
author={Zhaoyan Gong and Juan Li and Zhiqiang Liu and Lei Liang and Huajun Chen and Wen Zhang},
year={2025},
eprint={2509.03995},
archivePrefix={arXiv},
primaryClass={cs.CL},
url={https://arxiv.org/abs/2509.03995},
}