>Split up ingredients into more parts. I wrote a service called Zestful that does exactly this.[0] It's an interesting problem. It's hard to train an ML syste

>Split up ingredients into more parts. I wrote a service called Zestful that does exactly this.[0] It's an interesting problem. It's hard to train an ML system on the distinction between a preparation instruction like "finely chopped" and a note like, "just the white parts." My system treats "just the white parts" as a preparation instruction, but it throws away as irrelevant things like, "My favorite brand of black pepper is Smith's Fine Peppers." But I've looked at thousands of ingredients, and there are lots of wacky edge cases. I created the service by adapting an open source project the NY Times published and then abandoned.[1] They were trying to index all of their old recipes for what would eventually become cooking.nytimes.com. Their repo felt very experimental and half-baked, so it was interesting to turn it into a production service. I wrote about the process of creating the service on my blog.[2] [0] https://zestfuldata.com/demo https://zestfuldata.com/demo [1] https://open.blogs.nytimes.com/2015/04/09/extracting-structured-data-from-recipes-using-conditional-random-fields/ https://open.blogs.nytimes.com/2015/04/09/extracting-structu... [2] https://mtlynch.io/resurrecting-1/ https://mtlynch.io/resurrecting-1/