TindART was a visual arts recommender system I co-developed, published at ACM Multimedia 2020.

Co-developer & Co-author PythonPyTorchLightFMD3.jsRabbitMQ
2020

Highlights

  • Published research: Co-authored a paper at ACM Multimedia 2020 (A* conference) with the University of Amsterdam
  • Recommender system: Built a personal visual arts recommender that learns user taste through interaction
  • Visual analytics: Co-developed a comprehensive analytics tool to explore and understand recommendation patterns
  • Full system: Delivered an end-to-end system combining ML recommendations with an interactive frontend

TindART was a project at the University of Amsterdam where I was part of a team building a personal visual arts recommender system paired with a visual analytics tool. We worked under the guidance of PhD candidate Gjorgji Strezoski and professor Marcel Worring.

The system

The idea was to build a recommender that could learn a user’s taste in visual art through interaction and surface relevant works. The system combined recommendation algorithms (using LightFM for collaborative filtering) with a visual analytics interface built in D3.js that let users explore why certain works were being recommended. RabbitMQ and Celery handled the async processing pipeline.

The result

TindART like module showing a Rubens painting with thumbs up/down

The like module — swipe to teach the system your taste.

TindART visual analytics map showing user preferences across artworks

The analytics map — user preferences visualized across the collection.

The project led to a paper published at ACM Multimedia 2020, one of the top conferences in multimedia research. It was a meaningful intersection of the two things I care about: building real interactive systems and applying machine learning to solve a concrete problem.