Verification/Roadmap

Objectives

• Detect potential errors in order to correct them and prevent them
• Do not take into account the products whose information is questionable in the calculations of averages

Ideas to automatically detect products that contain errors entering information

Rules

• Sum of components of nutritional table well above 100g
• Sum of ingredients well over 100g (detection and complicated products )
• Confusion between "l" ( lowercase letter L ) and 1 (one) in the list of ingredients because of the OCR, which gives: nom_de_l'ingrédient l5 %
• Sum of " that sugar / starch / etc . " than carbohydrates (and the same for fat etc.).
• Energy over 4000 kJ
• Height from the upper portion to the size of package
• "CO2 Carbon Footprint" greater than 3 000g
• Over 12 additives
• Quantity of "sugars" greater than the amount of "carbohydrate"
• Sum of fatty acids greater than the amount of "lipids"
• Use the EU Organic codes to check whether the certification country is the same as the production country
• Year in a label is 2 or more greater than best before.
• etc.

Methods

• Consistency checks within a category
• Consistency checks with logical or biological rules
• Consistency checks with Laws
• Consistency checks with outside corpora
• Consistency checks of contributed data with OCR results

• Bots : several bots already exist.

Note

Each track (each rule) must have a unique number associated clearly: Once an accurate method of error detection is translated computer (is implemented), then this method becomes a rule. The automatically assigned number to this rule allows contributors to designate it unambiguously.