Case study: Hello, Cookie!

Hello, Cookie!
feat. Boris (1) + Sendhil (students, interfaces)
feat. Sendhil + team : schemas, roadmap

Overview

Goal, participants, purpose

Participants:

• A group gathering cookie recipes + turning them into collections

• A set of cookie data-sources (from old books, from people)

Goals:

• Identify users and use cases to get feedback + measure progress

• Structure schemas for each, define what parts of their use will touch R1 and other tools

• Identify similar approaches for comparison + benchmarking

Elements of a cookie recipe

• Atoms:

  • Ingredients

  • Measures

  • Tools

  • Actions

• Derived atoms:

  • Substitutes (equivalence classes of ingredients)

  • Definitions (aliases; concepts — "what is a chocolate chip cookie?”)

  • Categories (folksonomy of tags applied to the above)

• Recipes:

  • Recipe steps: combinations of atoms

  • Traditional recipes - sequences of steps + time + free description

  • Parametric recipes - models: tunable parameters + outcomes

• Types of categories:

  • by input

  • by output

  • by materials

  • related filters

    • filter by what is available

    • fuzzy filtering w/ substitutes

Sharing recipes

• Cookie book: 10 types, 100 recipes, 300 modules

• Overview: atlas/feature-universe

  • find recipes by feature,

  • find a specific recipe

  • stats on top recipe requests

• Recipe sharing

  • submit variant or new recipe

  • comment/review

Data sources

• Data Catalog

  • Index of sources from books, sites, scrapers

  • Find a maintainer for each, ask for more uniform provenance + metadata

• Glossaries and terms

• Existing recipes

  • Mining old cookbooks (OCR + NLP)

  • Scrapable websites: often limited structure + persistence

• Adjacent datasets :

  • In culture: Related music/art/books

  • For access: ingredient cost, nutrition, accessibility

  • For delight: taste graphs

• Meta datasets: Food ontology: foodkg

Process

• Creating schemas to match a set of sources

• Initial data entry by enthusiasts

  • Start by hand, with a spreadsheet for each schema / data source.

• Chef League: Major choices that guide taste prefs (meat v veg, sugar or no), modeling exceptions (binaries for allergies, cilantros, durian)
  

Proposed targets

• Collections w/ cookie recipes, definitions, and other data

  • Examples large (a cookbook) and small (all about one recipe)

• Initial data others can query, view, ingest. (by October)

• Articulate a contribution flow: how recipe writers, makers, testers add to collections

To incorporate

~ existing texts (Myhrvold team: scan+ocr; earlier work)
~ existing structured data : nutrition; diet substitutes; RDA
~~ existing ontologies: FoodOn, WD, FB. cuisines, KG connections
~ specialist data: cooking time; preservation-time in a fridge;
~ chef-specialties: techniques, reference works, taste clusters
~ categories: common classifiers (pescatarian, Atkins, Moroccan)
~ schemas: find/make a place to store + version schemas for food/cooking/recipes