So the menus rotate about once per month for most restaurants. And all of a sudden diners have to scroll through a ton of menus to find something they want. We can't use item-to-item collaborative filter because we always have a sparse data problem on the new menus.
So instead we are trying to create NLP embeddings on the menus and then ranking new menus based on similarity to old menus the diner liked.
So instead we are trying to create NLP embeddings on the menus and then ranking new menus based on similarity to old menus the diner liked.