DALK: Dynamic Co-Augmentation of LLMs and KG to answer Alzheimer’s Disease Questions with Scientific Literature
To fully run this project, one is recommended to install the requirement packages listed in requirements.txt.
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
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.
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, 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.