Toffee -- Topical Relevance Feedback Search

Iterative relevance feedback search using topic modeling and user feedback to guide the search towards the topics of interest.

More details can be found in:

Mikko Koho, Erkki Heino, Arttu Oksanen and Eero Hyvönen: Toffee - Semantic Media Search Using Topic Modeling and Relevance Feedback. Proceedings of the ISWC 2018 Posters & Demonstrations, Industry and Blue Sky Ideas Tracks, CEUR Workshop Proceedings, Monterey, California, USA, October, 2018. Vol 2180. Pre-print PDF

Requirements

• Hades for news corpus index in elasticsearch: https://github.com/SemanticComputing/hades
  • Elasticsearch uses the Hades esdata volume directly if present

Architecture

Running the initial training using the news corpus

docker-compose -f docker-compose-train.yml up

Running locally

Add a file named .env to the repository root with API_KEY= and the Google search API key. The REACT_APP_BACKEND should point to the web service address at the hosting server.

docker-compose build --build-arg 'REACT_APP_BACKEND=http://localhost:5000' frontend
docker-compose up -d

To deploy with several worker and prerender instances: (for docker version < 3.0)

docker-compose build --build-arg 'REACT_APP_BACKEND=http://localhost:5000' frontend
docker-compose up -d
docker-compose scale worker=3 prerender=3