UrbanKGent: A Unified Large Language Model Agent Framework for Urban Knowledge Graph Construction (PDF)

  

| **[1 Introduction](#introduction)** | **[2 Requirements](#requirements)** | **[3 Usage](#usage)** | **[4 Citation](#citation)** | **[5 Website](https://htmlpreview.github.io/?https://raw.githubusercontent.com/usail-hkust/UrbanKGent/main/UrbanKGent%20Demo/index.html)** |

1 Introduction

Official code for NeurIPS 2024 paper "UrbanKGent: A Unified Large Language Model Agent Framework for Urban Knowledge Graph Construction".

Urban knowledge graph has recently worked as an emerging building block to distill critical knowledge from multi-sourced urban data for diverse urban application scenarios. Despite its promising benefits, urban knowledge graph construction (UrbanKGC) still heavily relies on manual effort, hindering its potential advancement. This paper presents UrbanKGent, a unified large language model agent framework, for urban knowledge graph construction.

2 Requirements

python=3.8,torch=2.1.1, transformers=4.35.2, accelerate=0.25.0, geographiclib=2.0, request=2.31.0, geopy=2.4.1, geohash2=1.1

3 Usage

Agent Inference

To run OpenAI LLM agent, you need to set your key in main_TE/KGC.py:

python main_KGC/TE.py

To run the official Llama-2 LLM agent, you should put your LLMs model into the '/data/llm_models', and then open the Llama-2 inference local-host request:

python utils/open_llm_api.py
python main_KGC/TE.py

To run the fine-tuned Llama-2 LLM agent, you should put your LLMs model into the '/data/llm_models', and put your LLM adapter into the '/data/llm_model/', and then open the finetuned Llama-2 inference local-host request:

python utils/llama_lora_api.py
python main_KGC/TE.py

Agent Fine-tuning

To fine-tune tailored LLM agent, you need to prepare instruction set and put them into './sft_data/', and then finetune the model with LoRA architecture:

python finetune.py

The instruction tuning dataset is consist of trajectory generated in UrbanKGent Inferrence. We provide the example in './sft_data/', and you can construct tailored instruction tuning dataset with the similar format and process.

Our raw data, fine-tuned LLM adapter and constructed UrbanKG are available at "Google Cloud"

Agent Downloading

The opensource UrbanKGent-7b, UrbanKGent-8b and UrbanKGent-13b in this work could be downloaded from hugging face:

from huggingface_hub import snapshot_download

repo_id = "usail-hkust/UrbanKGent-13B"  # "usail-hkust/UrbanKGent-7B", "usail-hkust/UrbanKGent-8B"
local_dir = "./data/llm_models/UrbanKGent-13B"  # "./data/llm_models/UrbanKGent-7B", "./data/llm_models/UrbanKGent-8B"
local_dir_use_symlinks = False  #
token = "YOUR_KEY"  # hugging face access token

snapshot_download(
    repo_id=repo_id,
    local_dir=local_dir,
    local_dir_use_symlinks=local_dir_use_symlinks,
    token=token
)

4 Citation

If you find our work is useful for your research, please consider citing:

@article{ning2024urbankgent,
  title={UrbanKGent: A Unified Large Language Model Agent Framework for Urban Knowledge Graph Construction},
  author={Ning, Yansong and Liu, Hao},
  journal={Advances in Neural Information Processing Systems},
  volume={37},
  year={2024}
}

@article{ning2024uukg,
  title={UUKG: unified urban knowledge graph dataset for urban spatiotemporal prediction},
  author={Ning, Yansong and Liu, Hao and Wang, Hao and Zeng, Zhenyu and Xiong, Hui},
  journal={Advances in Neural Information Processing Systems},
  volume={36},
  year={2023}
}

5 Website

UrbanKGent is a online agent providing urban knowledge graph construction service for the researchers in the urban computing domain. You can visit our project website to utilize the deployed online UrbanKG construction service.