Skip to content

PyTorch module that generates a subgraph originating from the UMLS Knowledge Graph (KG) when presented with a biomedical question (from MedQA). It utilizes Language Models (LMs) to solve the next relation prediction task and graph node embedding to infer the tail given a head and the predicted relation, essentially performing link prediction.

Notifications You must be signed in to change notification settings

lorenzobalzani/neural-subgraph-retrieval

Repository files navigation

Neural Subgraph Retrieval

As a starting point, everything is supposed to be done inside a Docker container. Simply run the script launch_container.sh located inside the docker folder to build the image, if needed, and then launch the container.

End-to-end Subgraph Retrieval System

The complete subgraph retriever will soon be made available.

Separate Components

Predicting Next Relations

Data

The dataset designed for next relation prediction is open-source and can be downloaded from this link on the HuggingFace Hub.

Model

The model is open-source and accessible for download via this link on the HuggingFace Hub.

Link Prediction

Model

The model is open-source and accessible for download via this link on the HuggingFace Hub.

About

PyTorch module that generates a subgraph originating from the UMLS Knowledge Graph (KG) when presented with a biomedical question (from MedQA). It utilizes Language Models (LMs) to solve the next relation prediction task and graph node embedding to infer the tail given a head and the predicted relation, essentially performing link prediction.

Topics

Resources

Stars

Watchers

Forks