Taste Profile
A taste profile is a computational model of an individual's visual preferences, derived from their interactions with fashion items and used to generate personalised recommendations.
Origin & context
The concept extends from recommendation system research, where user preference models have been used in music, film, and e-commerce since the 1990s. Applied to fashion, it became more powerful with the development of vision-language embedding models capable of representing items in a shared semantic space — making it possible to compare the "style distance" between items a user likes and items they have not yet seen.
In fashion discovery
On Envie, a taste profile is a 768-dimensional vector — a point in an embedding space where visually similar items cluster. It is updated with every swipe using weighted averaging: likes attract, nopes repel, and recent interactions count more than older ones. It is different from a wishlist (which records conscious picks) because it captures patterns the user may not consciously recognise.
Examples
A user who consistently likes demi-fine gold jewelry with organic shapes and nopes silver and geometric forms will develop a taste profile that reliably retrieves more organic, warm-metal pieces — even from brands and categories they have never encountered before.
Related terms
Discover what your personal aesthetic actually is.
Request early access