/recipe170

The main code repository for '170' recipe analysis project.

Primary LanguagePythonMIT LicenseMIT

Recipe170

"Recipe170" is the title of our Global Recipe Analysis Project, which analyses the worldwide recipes with NLP-based solutions.

1. Quick start

  • 1.1 Environment specification

    • Miniconda (conda 23.11.0)
    • Python 3.9
    • Pipenv 2023.11.17

  • 1.2 Preprocessing

    • Step 1: cleaning texts
    • Step 2: unify quantity units
    • Step 3: get unique values and match all entries
    • Step 4: translate ingredients and quantities

2. TODOs

The overall processing of recipe data consists of a few steps:

  • Materials
    • a nice translator for Japanese--GPT-3.5-turbo API
    • parse materials + usage and export a mapping dataframe
    • parse materials and its upper-level materials, exporting a mapping dataframe
  • Units
    • numeric units, 1 0 0 g
    • textual units, 5、6個
    • enhanced units, 强弱
  • Procedure
    • longer-token translator
    • entity-recognition for cooking/timing/objects
  • Caveats
    • some recipes do not have ingredients!!!
    • create an error table for manual annotation (~20k from ingredient side)
    • use steps (ingredient extraction) to supplement to the ingredients

3. Issue Log

  • id=eb1d2e4604d93afd2753880c9f79b48e4d2fe582, issue=皮*小麦粉*3と3/4カップ
  • id=a00912dda86900e54e0df98fe658cb2f4686f23e, issue=皆さんのレシピ*みおりんさん＊8801、MIRELLEさん*8785、カヨリーヌさん*9176、まるりんさん*8799