Ingredients Extraction and Analysis

From Open Food Facts wiki
Revision as of 13:26, 13 May 2020 by Stephane (talk | contribs)
Jump to navigation Jump to search

This page describes how the ingredients list extraction and ingredients analysis is done on Open Food Facts and points to ressources that could be used to improve it.

Objectives

Ingredients list extraction

The goal of ingredients list extraction is to get the text of the ingredients list of each product in exactly the same form as it appears on the product package and label.

Ingredients analysis

Once the ingredients list is available, we need to analyze it to recognize the actual ingredients and indications of quantity, labels, processing etc. There is a lot of variety in how ingredients are listed on products, with many different synonyms, ways to indicate sub-ingredients etc.

The analysis needs to work for ingredients lists written in many different languages.

The output is structured data that links to our multilingual ingredients taxonomy.

Why it's important

Ingredients list extraction and analysis is necessary for many tasks:

  • Detecting food additives and allergens
  • Determining the degree of processing of food products (NOVA classification)
  • Identifying food products that can be or cannot be eaten by people following specific diets:
    • Vegetarian, vegan
    • Casher, Halal
    • Palm oil
  • Estimating the carbon impact of ingredients
  • Translating ingredients lists

High level view

Ingredients extraction and analysis.png


Ingredients list extraction

Data sources for ingredients lists

The possible input sources for the ingredients lists are:

  • Ingredients lists typed in by users
    • Time consuming and not pleasant task, especially on mobile
    • Can contain typos, but usually typed ingredients lists are very close to what is written on the product
  • Ingredients lists given by manufacturers in data files
    • Usually of very good quality, but depending on manufacturers, can contain typos and sometimes formatting errors
  • Photos of product labels
    • Photo quality varies a lot
      • Some products are hard to photograph (round cans and bottles, foil bags etc.)
      • Sometimes very poor lighting, orientation, camera, focus etc.
  • High resolution images or PDFs of the printable package
    • Available for a few producers
    • Perfect quality
    • Needs cropping and/or rotation to select the ingredients list

Steps for ingredients lists

Sample product for examples: https://fr.openfoodfacts.org/produit/5000112558265/coca-cola-zero

Picture taking

  • Taken with mobile app, uploaded to OFF server
Ingredients photo

Ingredients list cropping

  • Done on mobile app just after picture taking
    • Cropping may be very inaccurate
  • Or done on web site at a later time, possibly by another user
    • Cropping slightly easier than on mobile
Cropped ingredients

OCR

  • Launched after cropping, done through the server
  • Current solution:
    • Google Cloud Vision
    • Cloud Vision returns a JSON object which is stored on the server

Result:

Ingredients list cutting

  • The image sent by the OCR can also contain other text content
    • Things that are not ingredients
    • Ingredients in other languages
    • The word "Ingredients:"
  • Current solution
    • Hardcoded regular expressions
  • Other possible solutions
    • Language identification to remove other languages
  • Metrics
    • False negatives (words before or after the ingredients list that should have been removed)
    • False positives (words that were removed but are part of the ingredients and should have been kept)
      • It is very important to have as few false positives as possible as it destructs data
  • Test and training sets
    • Only a few adhoc tests run during builds
    • Test sets needs to be created

Validation and/or correction by users

  • Current solution:
    • Users on the app or the web site are shown the OCR result
    • OCR result is not applied if not validated by the user
    • but users tend to validate lists without changes even if there are errors, especially on mobile
  • Other possible solutions
    • Use the result of ingredient analysis to show users ingredients that were not recognized
    • Show spell suggestions

Spell correction

  • Current solution:
    • Currently only done during ingredients analysis, not during ingredients extraction
    • Very simple (and slow) implementation of Peter Norvig algorithm
  • Other possible solutions
    • Spell checkers trained on ingredients
      • Elastic search spellchecker
      • Simspell
  • Metrics
    • Recall and precision
  • Test and training sets
    • Language models can be build with lists of ingredients from OFF
      • e.g. including only ingredients lists from producers, or lists for which we have a very high ingredients recognition rate
    • Test sets need to be created
      • Run spellcheckers on actual ingredients lists from OFF, review corrections

Desired result:

  • "Eau gazéifiée ; colorant : E150d ; acidifiants : acide phosphorique, citrate de sodium ; édulcorants : aspartame, acésulfame-K ; arômes naturels (extraits végétaux), dont caféine."

Ingredients analysis

Ingredients taxonomy

Ingredients analysis is about matching the ingredient list to known ingredients in our multilingual ingredients taxonomy.

So a pre-requisite for good ingredients analysis is to have a comprehensive ingredients taxonomy that includes translations and synonyms in the target language.

The taxonomy is on GitHub (it is a very big file): https://github.com/openfoodfacts/openfoodfacts-server/blob/master/taxonomies/ingredients.txt

Ingredients processing taxonomy

A lot of ingredients list also contain information on how the ingredients have been processed (e.g. "cooked pork meat", "sliced tomatoes", "powdered garlic").

Instead of listing all possible combinations of processing for each ingredient in the ingredients taxonomy, we have created instead a taxonomy of processing methods that we use during ingredients parsing:

https://github.com/openfoodfacts/openfoodfacts-server/blob/master/taxonomies/ingredients_processing.txt

Steps for ingredients analysis

Ingredients pre-parsing

  • The ingredients list is transformed to make parsing easier
    • Remove / normalize strange characters
    • De-abbreviate abbreviations
    • Split enumerations
      • e.g. "Vitamins A, B et C" -> Vitamine A, Vitamine B, Vitamine C
    • Additives E-numbers normalization (E330, e330, e-330, INS 330, SIN330 etc.)
    • Additives classes + additive splits
      • e.g. "Colour caramel" -> Colour: Caramel
    • Split some "A of B, C and D" (but not all...)
      • e.g. "Huile de palme, colza et tournesol" -> Huile de palme, huile de colza, huile de tournesol
    • Handle * and other signs that indicate some ingredients are organic, fair trade etc.
      • e.g. "Pomme*, ..., *: ingrédient issu de l'agriculture biologique" -> "Pomme bio"
  • Current solution
    • Perl code and regular expressions
      • lib/ProductOpener/Ingredients.pm - preparse_ingredients_text()

Ingredients parsing

  • Separate individual ingredients and match them to the ingredients taxonomy
    • Extract properties of ingredients
      • Labels like organic, fair trade etc.
      • quantity (%)
      • processing (e.g. "cooked")
      • origin (e.g. "France")
    • Multi-level ingredients / sub-ingredients
      • e.g. "Fromage (Lait, présure, sel)"
    • Recognize when "A and B" is a single ingredient, or 2 ingredients
      • Uses the taxonomy to make the determination
  • Current solution
    • Perl code and regular expressions + multilingual ingredients taxonomy
      • lib/ProductOpener/Ingredients.pm - extract_ingredients_from_text()

Result:

ingredients: [
{
vegetarian: "yes",
text: "Eau gazéifiée",
id: "en:carbonated-water",
rank: 1,
vegan: "yes"
},
{
rank: 2,
id: "en:colour",
text: "colorant"
},
{
vegetarian: "yes",
id: "en:e150d",
rank: 3,
text: "e150d",
vegan: "yes"
},
{
text: "acidifiants",
rank: 4,
id: "en:acid"
},
{
vegan: "yes",
vegetarian: "yes",
rank: 5,
id: "en:e338",
text: "acide phosphorique"
},
{
text: "citrate de sodium",
rank: 6,
id: "en:sodium-citrate"
},
{
id: "en:sweetener",
rank: 7,
text: "édulcorants"
},
{
vegan: "yes",
id: "en:e951",
rank: 8,
text: "aspartame",
vegetarian: "yes"
},
{
vegetarian: "yes",
text: "acésulfame-K",
id: "en:e950",
rank: 9,
vegan: "yes"
},
{
vegetarian: "maybe",
rank: 10,
id: "en:natural-flavouring",
text: "arômes naturels",
vegan: "maybe"
},
{
id: "en:vegetable-extract",
rank: 11,
text: "extraits végétaux"
},
{
text: "dont caféine",
id: "en:caffeine",
rank: 12,
vegetarian: "yes",
vegan: "yes"
}
],

End to end metrics

Known and unknown ingredients

Results of further ingredient analysis

  • Number of products for which we are able to make a vegan / non-vegan or vegetarian / non-vegetarian determination
    • A non vegetarian/vegan ingredient triggers a non vegetarian/vegan result for the product
    • But to mark a product as vegetarian/vegan, we must have recognized all the ingredients of the product
    • https://fr.openfoodfacts.org/ingredients-analysis
      • France - Feb 8th 2020
        • Non végétalien 104789
        • Caractère végétalien inconnu 81966
        • Végétalien 27034
        • Peut-être végétalien 6029
  • Number of products for which we are able to make a NOVA determination
    • An ingredient marked as a NOVA 3 or NOVA 4 marker automatically makes the product NOVA 3 or 4
    • NOVA classification not made if too many ingredients are unknown (unless we have a NOVA 4 marker)
    • --> probably not a good indication of the quality of ingredients recognition

Ressources

Data

Ingredients taxonomy

How to improve ingredients analysis

This page describes How to improve ingredients analysis (after the text of the ingredients list has been extracted).