Brands
- These are the brands of the product.
- In the database, this field is called
brands. - See issues related to
brands. - See also: https://github.com/openfoodfacts/openfoodfacts-server/pull/4385
Help annotate brands using Hunger Games
Suggested solution for the taxonomy
Short-term solution
Make brands language-less
Iif a brand has a translated name, in some countries then we consider it to be a different brand (in a lot of cases the products sold under those translated names will be different).
For products that display the brand in only one language, we just record that brand, and we display it as-is.
Mid-term solution (to be defined/work in progress)
The taxonomy for brands should have following features:
- a single brand is defined by a block of text and are separated by white lines;
- use a four letters code separated by underscore - <iso_639_language_code> _<iso_3166_country_code> - for the key. Example: fr_be, de_at.
- a single line defines the brand in a specific language and country or an attribute. Each line starts with the key name followed by parameters and separated by a colon.
- a block can contain (mind the order):
- (optional) a reference to a parent defined by "<" + the key (as described above) and the first parameter of an existing brand. For example "<ab_cd:brand1" refers to the parent key "ab_cd:brand1, synonym_of_brand1". Remark: any attributes shared between the parent and the child only have to be specified in the parent, i.e., they are inherited by the child. Remark: if a reference to a parent is provided it should be the first line of the block.
- (required) a key (as described above). For example: "mn_op:brand 1". It should be unique, so that we can distinguish between brands with the same name.
- (optional) a default key (xx_xx:) to be used for any language that is not specifically listed. It is always xx_xx. Remark: if it is provided it should be the last line after all other languages.
- (optional) attributes - each brand can have one or more attributes. Remark: if provided attribute(s) should be the last line after all other languages and the default key (xx_xx):
- wikipedia:en: - the full link to a page, which explains the brand(!!!). Note that brands are not well available on wikipedia;
- wikidata:en: - the identifaction of the wikidata entry for the brand(!!). Note that brands are not well available on wikidata;
Example
de_de:lidl
de_au:lidl
fr_fr:lidl
fr_be:lidl
hr_hr:lidl
xx_xx:lidl
wikipedia:en:https://en.wikipedia.org/wiki/Lidl
wikidata:Q151954
<de_de:lidlde_de:snack day
xx_xx:lidl
In this example, we put de_de as first language because Lidl is a German company. Nevertheless, it should work the same starting with different language, at the condition that children refers to the parent accordingly.
Long-long-term solution (to be defined/work in progress)
Same as before, adding:
- (optional) attributes
- barcodeprefix:en: - the first 8(?) numbers of the barcode that belong to the brand. With this brands can be automatically assigned to a barcode. Also the barcodes of existing products can be checked;
- brand_owner_opencorporates:xx: - an identification of the probable brand owner/distributor on opencorporates. It is not always clear what should be written down here.
- idea: local customer service address (one for each country where the product is distributed, and a generic one)
- idea: eu trademark - maybe only useful for name trademarks? (not logo's). But not relevant to the consumer.
- idea: stores - the stores where the brand is sold. Maybe better to create a store taxonomy and link from there to brands?
Example
zz:laespanola
xx:La Española
barcodeprefix:en:8410226
barcodeprefix:en:8410660
category:en en:Olive tree products
brand_owner_opencorporates:ES:80245129
#94 products @2022-04-23
zz:latrappe
xx:La Trappe
barcodeprefix:en:8711406
category:en en:belgian-beers
label:en: en:authentic-trappist-product
website:nl:https://nl.latrappetrappist.com/nl/nl.html
brand_owner_opencorporates:BE:0838355558
wikidata:en:Q590921
#11 products @2022-04-23
Q&A
- Product has more than a single brands.
- Product is part of a bigger brand or group.
- Synonyms. Product is sold with 2 brands, each for different language/countries.
- For example, both
Nature ValleyandVal Naturecan be written on the same product, but not always. - Solution: make
Val Naturea child ofNature Valleyin this example becauseNature Valleyis the original U.S. brand (reference). - For example,
No NameandSans nomcan be written on the same product, but not always. - Solution: make
No Name,No Name Sans nomandSans nomthree separate children ofGeneric brandin this example because they are all lines ofGeneric brandwhich is a Canadian brand (reference). Or, the three of them as synonyms of each others and child ofGeneric brand. - For example,
Coop, in Switzerland, can add up to three brands on a product:Betty Bossi,KarmaandCoop. They justify it: "We offer a wide range of own-label brands and brand worlds." (source) - Solution:
- For example,
Liptonbelongs toUnileverbut this product is distributed byPepsico.Liptonused to belong toUnilever, and now belong toCVC Capital Partners.Lipton's ready to drink beverages belongs to bothUnileverandPepsiCo(who is distributor). - Solution:
- For example,
The Laughing CowandLa vache qui rit. - Solution: they both belong to the same parent (
Bel Group). They can be en_us and fr_fr. - For example,
Miko/wall's icecreams, calledFrigoin Spain butBreslerin Chile and Bolivia. - Solution: es_es:frigo and es_cl:bresler / es_bo:bresler will be on the same block
- For example, The Coca-Cola Company sells
Fantain Thailand underFantaandแฟนต้าnames. - For example,
Danonesells mineral water in Morroco under bothعين سايسandaïn Saïssnames. - For examples, Japanese/English (like
味の素/Ajinomoto) or Chinese/English (乐虎/Hi-Tiger) - Solution: th_th:แฟนต้า/fanta, แฟนต้า, fanta in the same block as en_us:fanta
- For example, both
- Also synonyms. Product brand has changed over time.
- Solution: use synonyms in the taxonomy.
- One brand is owned by two companies.
- For example,
Aldi NordandAldi Süduse the nameAldibut they operate in different areas (between countries, and inside Germany itself) - Solution: make
Aldi NordandAldi Südchildren ofAldi. In general, people will inputAldionly. However, if someone putAldi Nord, it should include its parent,Aldiautomatically.
- For example,
- The name of the brand is different from the name of the shop of these brand.
- For example
Aldi NordandAldi Südbrand, use the international (now unified) brandAldi(https://aldi.com/), which is also the common used shop name for bothAldibrands in Germany. So theAldi NordandAldi Südbrand, which are only used as shop indication brands on products (at least in Germany), have to resolved to the common shop nameAldiinstead ofAldi NordorAldi Süd. For the brandAldi Südthe shops could be also namedAldi Süd, as written on the logo used inside Germany, but as the shop name of theAldi Nordbrand was alwaysAldifrom the beginning, no one in Germany useAldi Südas name for these shops.
- For example
- One brand is distributed/produced locally by another brand
- For example, the licensing of beer brands: the same brand may be exploited under a licensing agreement by different companies depending on the country (and this can change with time). My understanding is that it is because for beer, it is a lot cheaper to produce locally than to transport internationally. In the UK San Miguel would probably be under Carlsberg Marstons, but in Spain it would be under Mahou San Miguel and in the Philippine or Hong Kong under the (original) San Miguel Brewery.
- For example,
Asahi, manufactured for the European market by its local subsidiaries (Peroniin Italy, for instance) but used to be licensed toCarlsberg MarstonsorAB Inbevdepending on the country. SomeAsahibrands are still licensed toAB InBevat least for some European markets. It used to be very opaque, but recent events made that obvious. - In Algeria, a lot of brands are done under licensing because of local laws which forbid a company which is not 51% owned by an Algerian to produce locally while import of a lot of products is forbidden or very limited.
- Solution:
- Same name shared by different brands.
- For example,
San Miguelis a beer in Spain [cervezas san miguel], food in Mexico [grupo agroindustrial san miguel], honey in Spain/France [Ramros Trading Company]) - For example,
Staris starfinefood in the U.S., Star S.p.A. in Italy, stardrinks in the United Arab Emirates, star from Ghadawat Indian airlines company, - For example,
Walkersrepresents 3 different brands (1, 2 and 3) - For example,
Doveis either cosmetic (from Unilever) or chocolates brand (from Mars) - Solution: In long term we could have some set of rules like if it is San Miguel and category is Beers, then rename Cerezas San Miguel. Same for food vs cosmetic.
- For example,
Cornettois a sub-brand of bothFrigoandMiko(see previous discussion) - Solution: make
Cornettochild of both brands.
- For example,
- Products without brands written.
- For example. No brand. I used the company name of the packager as brand to make search preview more useful.
- For example, unbranded product of the brand
Harvest House.The producer has a brandHarvest Houseused for other fresh tomato products, but it's products also appears under other brands.
- Brand spelling and formatting.
- Is it
Coop,coop,COOP,CO OP? - Solution: in taxonomy, there are no difference between lower and upper case, there are no difference between space and hyphen. We can make
coopa synonym ofco op. - Brands containing an apostrophe or quotes are often missing this one.
- Brands with special characters like &
- For example for &,
M&M's - Solution: special characters should be recognized. Eventually, use synonyms in the taxonomy.
- For example for &,
- Brands with commas
- For example for comma
Williams, West & Witt's Products - Solution: brand should be written without the comma. it is not possible to use commas in the tags (if you write comma it will start a new tag, this example will result in two tags:
WilliamsandWest & Witt's Productsor use ";" as separator as OpenStreetMap.
- For example for comma
- Brand with only numbers
- For examples,
1664,1883or365.
- For examples,
- Is it
- company structure - do we want to list (and research) all the relationships between owners, marketing companies, etc. I would suggest that we do NOT and limit ourselves to the brands and brand owner (and maybe the production company);
- comment: overkill - it is very tempting to add al kinds details for owners, etc. The actual owners of the product, brand, etc, do not seem useful to the consumer. We just should provide links to third parties for this kind information. I.e. wikipedia, wikidata,, opencorporates, ipo europe, ...
- Decision:
- parent brand - when should the parent brand be added? Only if the parent brand is available on the front of the packaging, or also when it is shown on the back of the packaging, or when we can find out the legal final parent owner of a brand? This choice might have an implication for how we structure the data.
- comment @benbenben: at least having sub-brands/lines of a brand would be beneficial - for quality checks as well as monitoring number of products from the same distributor - even if the brand is not written on the package.
- For example,
S-budgetis a child ofSpar. - For example,
Pilosis a child ofLidl.
- For example,
- Decision:
- comment @benbenben: at least having sub-brands/lines of a brand would be beneficial - for quality checks as well as monitoring number of products from the same distributor - even if the brand is not written on the package.
- EAN manufacturer codes - is there an open database which we could use?
Observations summary
Summarizing the observations note above, we see the following brands:
- 1 universal brand, exact same name used in all countries and languages. e.g. "Nutella"
- 2 brand that is translated in different languages or scripts "The Laughing Cow", "La vache qui rit" (example) EDIT they both belong to the same parent (Bel Group)
- 3 brands that have the same name, but used in different languages
- 4 brands that have the same name, but used in different countries
- 5 brands that have the same name, and used in the same country. (e.g. "Ferrero" in Italy: there's also a pasta brand. EDIT pasta brand in Italy named Industria-alimentare-ferraro).
- 6 brands in non-latin scripts, which can not be latinised
- 7 parent brands are sometimes shown on packaging
Use cases (UC below)
The brands taxonomy has multiple applications within Open Food Facts. These are:
- 1 Display the brands of a product, in the language requested by the user;
- 2 Have a way to list all products of a brand;
- 3 Let users enter brands for a product, as they appear on the package (as free text);
- 4 Let the user select the correct brand from a list of existing brands. If the same brand text occurs multiple times, the user must be able to select the applicable one;
- 5 Suggests a brand to the user based on the manufacturer part of the barcode and other information;
- 6 Infer category and labels from brand - some brands are only used for specific products. This implies that the product category and possible labels can be implies (suggested);
- 7 Infer brand from manufacturer code within the barcode;
- 8 Barcode/Brand quality check - if the manufacturer part of the barcode does not match the specified brand, there is an error in either of them;
Design considerations
The observations and use cases lead to several design considerations:
- Unique brand key - as the same brand (string) can exist in multiple geographic areas or within the same geographic area, there must be a way to uniquely distinguish between the various brands. Otherwise, a user can not enter the correct brand (UC1), nor can we list all brands (UC2);
- Uses selectable brand - there must be a brand name in the same language/script of the package. If that name occurs multiple times in the taxonomy, it must be specified by product category, country sold, etc in order to make it unique. For instance, the label Taste occurs in France and Argentina, but is used for different categories. So the user should have the choice Taste (category 1 - France) and Taste (category 2 - Argentina). Maybe this can be mixed with the key. (UC4). EDIT: there are no Taste brands neither in France not in Argentina.
- Language/script specific brands - a way to code a single brand in multiple scripts and/or languages, for example in Chinese, Arabic and English (UC1)
- Language independent brand - a way to encode a brand that is valid for multiple languages (UC1);
Brands and Robotoff (AI)
Preventing OFF AI to detect some false brands: see brand_taxonomy_blacklist.txt.
Help to collect brands
The AI of Open Food Facts, called Robotoff, is trying to identify brands in several ways. We use OCR in combination with known values. We also have another technique using pure computer vision and annotations to create cluster of brands by visual similarity. The annotations made by Robotoff are provided to users, asking them to answer to a simple question. There are also used by Hunger Game. Everyone can use Hunger Game, but be careful to the issues mentioned in this current page.
- on 2020-10-29, there were 43200 annotations and 22748 remaining
- on 2020-11-17, there were 46911 annotations and 22720 remaining
- on 2020-11-17, there were ? and 57879 remaining
Known challenges
Some brand names are related to common name, expressions or ingredients producing false positive with AI tool. For examples:
- Racines (means "roots" in French)
- La Truffe (means "the truffle" in French)
- Pure Protein
- Best Choice is a brand, not a tagline
- Great Value also
- The Belgian, often entered
Belgian; it produces many false positive from our AI.