Difference between revisions of "Showstoppers for Ingredients Extraction and Analysis"
Jump to navigation
Jump to search
(Created page with "=== Step 1 - Remove pet food & Cosmetics === * AI: identify them, especially with ingredients, and move them to OBF, OPFF, OPF ** https://hunger.openfoodfacts.org/questions?ty...") |
|||
Line 7: | Line 7: | ||
=== Step 2 - Fix ingredients stored in the wrong language === | === Step 2 - Fix ingredients stored in the wrong language === | ||
* AI: automatic image deselection & reselection | * AI: automatic image deselection & reselection | ||
+ | ** [https://github.com/openfoodfacts/robotoff/discussions/1192 Add a question to propose unselecting ingredient image if no ingredients are detected] | ||
+ | ** [https://github.com/openfoodfacts/robotoff/discussions/1214 Compare ingredient images and current ingredient lists] | ||
* AI: drastic cleanup of bad lists | * AI: drastic cleanup of bad lists | ||
* AI: Data quality facets about ingredients become errors | * AI: Data quality facets about ingredients become errors | ||
=== Step 3 - Fix errors of ingredients cutting === | === Step 3 - Fix errors of ingredients cutting === | ||
+ | * Crop/Recrop more tightly: [https://github.com/openfoodfacts/openfoodfacts-ai/issues/40 Autocrop ingredient lists] | ||
* Add more cutters in more languages | * Add more cutters in more languages | ||
* Add more cutters in French | * Add more cutters in French | ||
+ | * Move it to AI/LLM: [https://github.com/openfoodfacts/openfoodfacts-ai/issues/242 Extract automatically ingredient list from OCR] | ||
=== Step 4 - Fix errors with Bad OCRs === | === Step 4 - Fix errors with Bad OCRs === | ||
* Reprocess OCRs | * Reprocess OCRs | ||
* Perform spellcheck | * Perform spellcheck | ||
+ | ** [https://github.com/openfoodfacts/openfoodfacts-ai/issues/314 Use large language models (LLM) to perform ingredient list spellcheck] | ||
=== Step 5 - Bad splitting into individual ingredients === | === Step 5 - Bad splitting into individual ingredients === | ||
* Add a multilingual no cut list of words like U.H.T Cat 1… | * Add a multilingual no cut list of words like U.H.T Cat 1… |
Revision as of 15:26, 17 January 2025
Step 1 - Remove pet food & Cosmetics
- AI: identify them, especially with ingredients, and move them to OBF, OPFF, OPF
Step 2 - Fix ingredients stored in the wrong language
- AI: automatic image deselection & reselection
- AI: drastic cleanup of bad lists
- AI: Data quality facets about ingredients become errors
Step 3 - Fix errors of ingredients cutting
- Crop/Recrop more tightly: Autocrop ingredient lists
- Add more cutters in more languages
- Add more cutters in French
- Move it to AI/LLM: Extract automatically ingredient list from OCR
Step 4 - Fix errors with Bad OCRs
- Reprocess OCRs
- Perform spellcheck
Step 5 - Bad splitting into individual ingredients
- Add a multilingual no cut list of words like U.H.T Cat 1…
Step 6 - Bad ingredient properties recognition
- Processing, Origins, Other properties
Step 7 - Missing translations or synonyms
- Translate the taxonomies using LLM recipes
- Add synonyms
Step 8 - Truly untaxonomized ingredients
- Add those to the taxonomy
Step 9 - Missing or Bad CIQUAL associations
- Add CIQUAL matches
- Add matches from other Nutrition databases
Step 10 - Bad or insufficient categorization
- Identify problematic categories
- Play Hunger Games on problematic categories
Step 11 - Bad nutrition
- Clean nutrition
Step 12 - Additional data not the ingredient list
- Origins, -25% de sel, …