Research possible datasets to use

[WagnerMarcel]

GanttStart: 2022-01-01 GanttDue: 2022-01-16 Research possible large datasets of recipes which can be used in order to train the NN. The datasets should contain atleast 10k entries in order to prevent overfitting.

[WagnerMarcel]

Each analysis of dataset should contain:

• Plantuml diagram of json structure
• Amount of recipes
• Plots:
  • Num of Instructions
  • Num of Ingredients
  • Most used words in Instructions
  • Most used words in Ingedients
  • (Optional) Kind of dish

[WagnerMarcel]

Use Recipe1M

Only use recipes with no more than 20 ingredients and 30 instructions

Output data format:

[
  {
    "title": "NULL",
    "ingredients": [
      {
        "ingredient": "NULL",
        "amount": "NULL",
        "unit":"NULL"
        "instruction": "NULL" (optional, left out in first approach)
      }
    ],
   "instructions": [
    {
      "instruction":"NULL"
    }
   ]
  }
]

Formatted with: https://jsonformatter.org

[WagnerMarcel]

Recipe box

Data Structure of recipeBox[^1]:

Property	Amount
Recipes (stripped by empty ones)	124647
Ingredients (total)	1316950
Ingredients (unique words[^2])	22364
Numeric expressions in ingredients	1369071
Instructions (total split by ".")	1924048
Instructions (unique words)	40387
Numeric expressions in instructions	913097

Words in Ingredients

With numeric expressions

Without numeric expressions

Words in Instructions

With numeric expressions

Without numeric expressions

Initial Parsing

Separating units and ingredients:

[^1]: Hashed strings can be replaced with cleartext by running the scraper again. [^2]: Stripped by unique chars (",.*®©™()[]") and transformed to lowercase. Still ~10k only appear once.

[mscholl96]

Recipe1M

https://github.com/torralba-lab/im2recipe-Pytorch

Structure of data

Property	Amount
Recipes	1029720
Number of ingredients[^1]	9605394
Different unique words in ingredients	91803
Numeric expressions in ingredients[^2]	10582774
Number of instructions[^1]	10767598
Different unique words in instructions	254082
Numeric expressions in instructions[^2]	4187244

[^1]: String which can consist of several words, see PlantUML [^2]: Describing amounts such as "1 apple"or "1/2 apple". These numeric values are kept in one separate category.

Words in ingredients

With numeric expressions

Without numeric expressions

Words in instructions

With numeric expressions

Without numeric expressions

[WagnerMarcel]

Analysis of duplications by title

Property	Amount
Titles in recipeBox (total)	124595
Titles in recipeBox (unique)	114281
Titles in recipe1M (total)	1029720
Titles in recipe1M (unique)	809994
Titles in openRecipes (total)	173278
Titles in openRecipes (unique)	151025
Duplicated titles recipeBox / recipe1M	74995
Duplicated titles openRecipes / recipe1M	102108
Duplicated titles recipeBox / openRecipes	33102

[WagnerMarcel]

OpenRecipes

Structure of data

Property	Amount
Recipes	173278
Number of ingredients	1693953
Different unique words in ingredients	27454
Numeric expressions in ingredients	2005536
Number of instructions	0
Different unique words in instructions	0
Numeric expressions in instructions	0

Words in ingredients

With numerical expressions

Without numerical expressions

[WagnerMarcel]

NowYoureCooking

Structure of data[^1]

Property	Amount
Recipes	28287
Number of ingredients	284307
Different unique words in ingredients	77136
Numeric expressions in ingredients	236035
Number of instructions	232211
Different unique words in instructions	24442
Numeric expressions in instructions	88687

Words in ingredients

With numeric expressions

Without numeric expressions

Words in instructions

With numeric expressions

Without numeric expressions

[^1]: Approximated to json data format, recipes are provided in a custom dataformat.