OnSET

Live Demo

In progress...

Description

OnSET should enable a more explorative ontology querying experience for non-expert users, and to build a quick understanding of the scale and nature of a KG that is in use. The toolkit consists of multiple smaller systems, complimenting each other. We have three default views ready:

• a query view; where users can construct queries either from topics or from natural language prompts,
• the explorative view; enabling users to view the ontology as a whole, view sparse relations and explore the named instances, and
• the linked view; showing users the hierarchical ontology and topic model at the same time, creating a rich visualization of multiple dimensions within the ontology.

Setup instruction

1. Install Apache Fuseki such that you have a working riot command line tool (https://jena.apache.org/documentation/fuseki2/)
2. Setup a local qlever instance (https://github.com/ad-freiburg/qlever-control) (the used configurations can be found in docker/bto-data and docker/dbpedia-data), you should be able to just run from the docker/*-data-directories:

qlever get-data
qlever index
qlever start

3. Start auxiliary DB using cd docker/onset-data && docker-compose up -d
4. Backend setup:
   • install uv (https://docs.astral.sh/uv/getting-started/installation/)
   • install dependencies using CMAKE_ARGS="-DGGML_CUDA=on" uv sync (or CMAKE_ARGS="-DGGML_METAL=on" uv sync on Mac)
   • start backend (should initialize DB on first start) using the correct Python installation using python -m uvicorn api:app --reload --port 8001
5. Frontend setup:
   • install a somewhat recent Node version
   • install dependencies by first cd frontend && npm i -D
   • then start the frontend using npm run dev

• Profit?