DALK: Dynamic Co-Augmentation of LLMs and KG to answer Alzheimer’s Disease Questions with Scientific Literature

Update

10.17: DALK has been accepted by EMNLP 2024 Findings! Check our camera-ready version here.

Overview

Dependency

To fully run this project, one is recommended to install the requirement packages listed in requirements.txt.

LLMs for KG

To extract AD-KG for the following steps, one need to first replace the following parts in LLM4KG/api_utils.py:

palm.configure(api_key='')
api_key = ''

Then run LLM4KG/llm2kg_s2s.py or LLM4KG/llm2kg.py to extract AD-KG in generative method or RE method.

We also provide the AD-KG data directly in KG4LLM/Alzheimers/train_s2s.txt and KG4LLM/Alzheimers/train.txt

KG for LLMs

ADQA Benchmark

We put the samples of ADQA benchmark in KG4LLM/Alzheimers/result_filter/ for the original data and KG4LLM/Alzheimers/result_ner for the data with extracted entities.

Augment LLMs with AD-KG

To augment LLMs' inference with AD-KG, one need to first create a Blank Sandbox on Neo4j, and replace the follow parts in KG4LLM/MindMap_revised.py

YOUR_OPENAI_KEY = ''
uri = ""
username = ""
password = ""

And then, build a folder under Alzheimers called result_chatgpt_mindmap and run

cd KG4LLM
python MindMap.py

It may take a while to upload the local AD-KG data to neo4j.

Credits: This work began as a fork of MindMap' repository: MindMap. If you found our code useful, please consider citing:

@article{li2024dalk,
  title={DALK: Dynamic Co-Augmentation of LLMs and KG to answer Alzheimer's Disease Questions with Scientific Literature},
  author={Li, Dawei and Yang, Shu and Tan, Zhen and Baik, Jae Young and Yun, Sunkwon and Lee, Joseph and Chacko, Aaron and Hou, Bojian and Duong-Tran, Duy and Ding, Ying and others},
  journal={arXiv preprint arXiv:2405.04819},
  year={2024}
}