Open aleene opened 2 weeks ago
Robotoff is actually already leveraging the ingredient list (among other signals) to predict the category. We're using:
as signals. We could advertise how the prediction is done discreetly in the UI. cc @raphael0202
That is great. So my cleaning up will help the predictions as well.
As we are now adding additional info on related categories and ingredients into the taxonomy, this could be added as well as signal??
Problem
Many products do not have an associated category. This makes it impossible to calculated a Nutriscore.
Proposed solution
Use the ingredients to suggest an associated category.
Several categories have expected ingredients listed in the categories taxonomy. If a product without a category, but with ingredients, matches these ingredients, a suggestion can be made.
This can be applied to all products that have a single ingredient, such as fruits, vegetable, peanuts, milks, flours, oils, etc. It might be possible to extend this to products with multiple ingredients, such as cheeses.
Number of products impacted
50% of the products?