This is the official github repo of Think-on-Graph (ICLR 2024). If you are interested in our work or willing to join our research team in Shenzhen, please feel free to contact us by email (xuchengjin@idea.edu.cn)
ToG The code for paper: "Think-on-Graph: Deep and Responsible Reasoning of Large Language Model on Knowledge Graph". Our paper is accepted by ICLR 2024 🥳🥳🥳. Here is the illustration of ToG: The pipeline of ToG: Project Structure : Pip environment file. : Pre-defined Wikidata APIs, copy from . : The main file of ToG where Freebase as KG source. : Same as above but using Wikidata as KG source. : The prompts for the ToG to pruning, reasoning and generating. Get started Before running
ToG, please ensure that you have successfully installed either **Freebase** or **Wikidata** on your local machine. The comprehensive installation instructions and necessary configuration details can be found in the file located within the respective folder. The required libraries for running ToG can be found in . When using the Wikidata service, copy the and files from the directory into the folder. md How to eval Upon obtaining the result file, such as , you should using the script from the directory to
convert the to . Then, evaluate using the script in the folder (see in folder). How to cite If you interested or inspired by this work, you can cite us by: Experiment: Application: Claims This project uses the Apache 2. The project assumes no legal responsibility for any of the model's output and will not be held liable for any damages that may result from the use of the resources and output. FYI We are looking for self-motivated interns at
IDEA (Shenzhen). If you are interested in the topics of LLMs and KGs, please send us your resume by email.
常见问题
记录了多少星标和分叉？
DataArcTech/ToG 有 663 个星标和 74 个分叉。
仓库排名是多少？
DataArcTech/ToG 排名第 15,308。
记录的语言和类别是什么？
DataArcTech/ToG 有 663 个星标，语言为 Python，类别为 AI Engineering。
data/: Evaluation datasets. See data/README.md for details.
CoT/: CoT methods. See CoT/README.md for details.
eval/: Evaluation script. See eval/README.md for details.
Freebase/: Freebase environment setting. See Freebase/README.md for details.
Wikidata/: Wikidata environment setting. See Wikidata/README.md for details.
tools/: Common tools used in ToG. See tools/README.md for details.
ToG/: Source codes.
client.py: Pre-defined Wikidata APIs, copy from Wikidata/.
server_urls.txt: Wikidata server urls, copy from Wikidata/.
main_freebase.py: The main file of ToG where Freebase as KG source. See README.md for details.
main_wiki.py: Same as above but using Wikidata as KG source. See README.md for details.
prompt_list.py: The prompts for the ToG to pruning, reasoning and generating.
freebase_func.py: All the functions used in main_freebase.py.
wiki_func.py: All the functions used in main_wiki.py.
utils.py: All the functions used in ToG.
Get started
Before running ToG, please ensure that you have successfully installed either Freebase or Wikidata on your local machine. The comprehensive installation instructions and necessary configuration details can be found in the README.md file located within the respective folder.
The required libraries for running ToG can be found in requirements.txt.
When using the Wikidata service, copy the client.py and server_urls.txt files from the Wikidata directory into the ToG folder.
Upon obtaining the result file, such as ToG_cwq.jsonl, you should using the jsonl2json.py script from the tools directory to convert the ToG_cwq.jsonl to ToG_cwq.json. Then, evaluate using the script in the eval folder (see README.md in eval folder).
How to cite
If you interested or inspired by this work, you can cite us by:
@misc{sun2023thinkongraph,
title={Think-on-Graph: Deep and Responsible Reasoning of Large Language Model with Knowledge Graph},
author={Jiashuo Sun and Chengjin Xu and Lumingyuan Tang and Saizhuo Wang and Chen Lin and Yeyun Gong and Heung-Yeung Shum and Jian Guo},
year={2023},
eprint={2307.07697},
archivePrefix={arXiv},
primaryClass={cs.CL}
}
Experiment:
Application:
Claims
This project uses the Apache 2.0 protocol. The project assumes no legal responsibility for any of the model's output and will not be held liable for any damages that may result from the use of the resources and output.
FYI
We are looking for self-motivated interns at IDEA (Shenzhen). If you are interested in the topics of LLMs and KGs, please send us your resume by email. Our email address is xuchengjin@idea.edu.cn
