The problem with this algorithm is that it only provides you with ingredients, and not with cooking instructions. It's not hard to gather data from all the recipes in the world, find which ingredients each recipe has, and find which recipes are successful, and then combine compatible ingredients from successful recipes together. What's really hard it so provide the cooking instructions, and not just the ingredients.
By the way, as a programmer myself, I read the paper and I discovered that it definitely doesn't take cooking instructions into consideration - perhaps that's why they didn't, in their paper, provide any mix of ingredients that includes meat.
PS: This same concept can be applied, perhaps with more success, in the medical field. In fact, it was the topic of an R&D paper we submitted last year.