Wikidata

Wikidata:Property proposal/safety classification and labelling

< Wikidata:Property proposal

safety classification and labelling edit

Originally proposed at Wikidata:Property proposal/Natural science

Descriptionclassification and labelling data for risk identification about chemicals
Data typeItem
Domainchemical substance (Q79529)
Allowed valuesclassification scheme (Q5962346)
Example * see qualifiers below
SourceSame as for GHS system, UN regulations or NFPA 704
Planned useSame as for GHS system, UN regulations or NFPA 704
See also

Currently several classification and labelling systems are used to identify the risks associated to chemicals. The most important ones are the GHS system, NFPA 704 or the UN Recommendations on the Transport of Dangerous Goods used for ADR/RID, IATA and IMDG transport. These systems require a complete set of data and the data set can be different depending on the sources or the evaluation date. Currently different properties already exists in WD to create the data set but no way exists to retrieve directly all relevant data of one data set from the item and some filters based on reference data need to be applied. The goal of the new property is to group all data from one source into a single structure. This property will be used for different classification and labelling systems. See examples below. Snipre (talk) 00:18, 29 December 2017 (UTC)[reply]

Examples
dichlorine (Q1904422):
safety classification and labelling:Globally Harmonized System of Classification and Labelling of Chemicals (Q899146)
P728 (P728): H270, H280, H330, H314, H318, H335, H400
GHS signal word (P1033): danger
P940 (P940): P202, P244, P260, P264, P271+P403, P273, P280+P284, P370+P376, P405, P501
icon (P2910): comburant
icon (P2910): gas under pressure
icon (P2910): corrosive
icon (P2910): toxic
icon (P2910): pollutant
safety classification and labelling:NFPA 704 (Q208273)
NFPA Health (P993): 4
NFPA Fire (P994): 0
NFPA Instability (P995): 0
NFPA Special (P877): OX
safety classification and labelling:UN Recommendations on the Transport of Dangerous Goods (Q7865437)
UN number (P695): 1017
Kemler code (P700): 268
UN class (P874): 2
UN classification code (P875): 2TOC
UN packaging group (P876): novalue
icon (P2910): 2.3
icon (P2910): 5.1
icon (P2910): 8

Open questions:

  Notified participants of WikiProject Chemistry

Discussion
TomT0m proposal describing classification method for GHS
(hiding for more clarity in this discussion. — Wostr)
  • Counter proposal : create items for danger classes, like « H336 » or « H200 » for the first classification scheme, and use the pictogram property once on each of these new items.
    here is what those new items would look like :
    ⟨ Explosif instable (fr)/unstable explosive (en)/… ⟩ Globally Harmonized System of Classification and Labelling of Chemicals (Q899146) code [[Special:Search/Property:Globally Harmonized System of Classification and Labelling of Chemicals (Q899146) code|Search]] ⟨ H200 ⟩
    ⟨ unstable explosive ⟩ instance of (P31)   ⟨ GHS hazard class ⟩
    (metaclassification)
    ⟨ unstable explosive ⟩ pictogram Search ⟨ relevant icon ⟩
    /
    ⟨ unstable explosive ⟩ pictogram Search ⟨ relevant icon ⟩
    / … (Then to retrieve the set of picture it amounts to a query (or a path expression) : select ?icon where { ?hazard_class instance of <GHS hazard class> . Qmixture_or_compound <subclass of>* ?hazard_class . ?hazard_class ?icon (assuming we use « subclass of», see below)
    A benefit of this approach is that instead of relying on external knowledge to interpret the meaning of the codes, we will directly encode some of the meaning into Wikidata statements, and even encode parts of the documents like http://www.unece.org/fileadmin/DAM/trans/danger/publi/ghs/ghs_rev07/French/ST-SG-AC10-30-Rev7f.pdf into Wikidata. Especially, we can encode relationships from combined hazard classes with the base class that are combined, or relationship between hazard classes (I assume H220 (Gaz extrêmement inflammable) H221 (Gaz inflammable) are linked and that H220 is a subclass of H221). It will also be possible to link these classes with preexisting Wikidata classes - I assume the gaz class is a subclass of fuel gas (Q909363).
    Theorically, as all of this is a matter of classification, it’s enough to classify a mixture using subclass of (P279) for example to retrieve all the relevant informations and automatically extract the relevant information from here. This would actually make this proposed property useless and would actually be a lot more economic in statements, as for example classify a substance as a subclass of the H221-class would imply it is a gaz for free, would allow to retrieve the associated codes and picture with simple queries. The expense would be a little work to encode the norms in Wikidata. author  TomT0m / talk page 11:06, 4 January 2018 (UTC)[reply]
    @TomT0m: As you are not using proper names for GHS labelling elements, I can't be sure that I understand your proposal correctly. There are four elements of GHS labelling:
    • GHS pictograms represented by icon and name/code (eg. GHS01 is 'exploding bomb' and the respective pictogram is File:GHS-pictogram-explos.svg)
    • signal word: 'danger' or 'warning'
    • H-phrases (and EUH-phrases) represented by code and descritpion (like 'H200' : 'Unstable explosive')
    • P-phrases (like above, e.g. 'P103' : 'Read label before use'
    I think that you trying to combine two things in this proposal: GHS classification and GHS labelling. I have to write it again, but GHS data in WD (and in Wikipedias) ARE NOT CLASSIFICATION only labelling. Therefore, no information on the classification of substances according to the GHS criteria can be derived from this data. GHS classification has two elements:
    • GHS class and category (e.g. 'STOT SE 1' for specific target organ toxicity for single exposure, category 1)
    • H-phrases assigned to GHS class and category (but these H-phrases are sometimes different from H-phrases used in labelling)
    So adding classificton would require additional properties. What's more:
    • about first sentence of your conter-proposal: H-phrases and GHS pictograms are equal elements of labelling, you can't assign GHS pictograms to H-phrases in a way that would allow retrieving any useful information.
    • Your assumption that H220 is a subclass of H221 (or any other H-phrase) is not correct. In a WD H-phrases can only be instances of GHS hazard statement (Q28360). There are no relations between phrases that would allow hierarchisation or something.
    • Your assumption with fuel gas (Q909363) is also wrong. I don't think that all substances having H220/H221 are fuel gas (Q909363). The only conditions are those specified in GHS, so fuel gas (Q909363) may be related term to some substances in GHS class, but I see no condition in GHS that every substance classified as flammable gas has to be fuel gas (Q909363).
    • Furthermore, I really don't know how we would choose P-phrases from many possibilities. Do I have to tell that in EU labelling usually consist of up to 5 P-phrases, which are chosen from many available phrases by manufacturer/importer etc.?
    I cannot agree with any part of your counter-proposal if I understand it correctly. We are gathering data here and we cannot act like classifiaction/labelling body, I think that some parts of your proposal would lead to WP:OR. Wostr (talk) 13:18, 4 January 2018 (UTC)[reply]
    @Wostr: : Essentially, may point is : firts, « GHS class » are specific case of classification. Wikidata classes can handle this classification as any other classification. Second, when you know the GHS class, you know the H-phrases and P-phrases. Which entails that if you got a link from the (wikidata) substance item to the (wikidata) GHS class item, assuming the GHS class item (or some parent class of it) contains the H-phrases and P-phrases, it’s redundant and useless to repeat it in the substance item. There may be a case if sometimes phrases are added for some substance - actually we have already cases in Wikidata to handle this kind of stuffs, like does not have part (P3113).
    « Your assumption that H220 is a subclass of H221 (or any other H-phrase) is not correct. » This is not exactly put that way. I assume there is a class (that can be defined, or that exists already in ) in which all substance with an H220 phrase is classified, same for H221, and that there is a relationship between those classes.
    There is probably an interwiki conflict on fuel gas (Q909363) as the frwiki article linked to this item is about flammable gaz. Which is not the same as fuel (@Snipre: does that semm this correct ?). So my bad, assume I was referring to a Wikidata item about « flammable gaz ».
    On original work : I don’t think so. There is a large body of literature explaining and defining the notions at sake, this is probably just a matter of extracting the relevant bodies of the documents that describe the norm and writing this in term of Wikidata vocabulary, just a factual translation. If this is original work, then most of Wikidata is. The point is not to copy over databases field by field in Wikidata. It’s to represent existing datas the Wikidata way, which we decide. There is no policy that we should use only preexisting database scheme or ontology, that would not even be possible as the Wikidata data model is unique. Wikidata as a consequence requires at least rewriting to some extent to rewrite the datas to fit into the model we choose (carefully) all together, just as rewriting into another language. The only requirement is that there is relevant sources to statements. author  TomT0m / talk page 13:59, 4 January 2018 (UTC)[reply]
    @TomT0m:, Second, when you know the GHS class, you know the H-phrases and P-phrases – that's not true for EU GHS (maybe for OSHA version it is, I don't know). You have (a) classification H-phrases (and you have to apply some rules [e.g. if A and B, show only B] to know which H-phrases should be used in labelling, because some elements that comes from classification are omitted when it comes to label specific substance), (b) you have a set of all possible P-phrases – you don't know which phrases to use in labelling (this is a decision of authorised entities like manufacturers etc.). Also, how can you tell from classification if specific compound has to be label with EUH phrase? Simply: you can't.
    Okay, let's put it that way: assumptions about H220/H221 are not correct because it is a result of classification — you can connect many H-phrases to specific GHS classes (but not to categories) [however, in some cases that would not be so easy, as e.g. H271 may be present in Oxidizing liquids or in Oxidizing solids]. But if you want GHS classification in WD, you would have to create new properties for GHS classes and GHS categories. From H-phrases you cannot derive the full information about classification. If you have those new properties for GHS class/category and a set of items about specific classes/categories, then you can connect them to some existing items, but it will mean only that e.g. for oxidising liquids class that some oxidisers in liquid state are classified as oxidising liqids, but some may not be (because they do not meet specific GHS criteria). Such connection between GHS classes and existing items will have very limited use.
    About OR: you can't derive labelling data from classificaton data by yourself (or by an algorithm or something). Such procedure is reserved for authorized entities. In other words: if you have classification data in GHS you have only classification data and nothing else. If you also want to have labelling data – you have to add this data separately. Because you have no formal rights to translate GHS classification into GHS labelling (doing so would be a clear example of original resarch). But think fo it: why Wikipedias don't use such trick in infoboxes if it's possible – adding only classification and labelling would be displayed based on some algorithm written in Lua or even using parser functions? Because labelling obtained in such way wouldn't be correct and wouldn't have any formal basis.
    Last but not least, there is a huge difference between GHS labelling and GHS classification in terms of potential usage of such data in WD. GHS labelling may be used by Wikipedia as well as other databases/webpages that provides informations for consumers (that means all end-users and not manufacturers, importers etc.). But GHS classification IMHO has no practical use anywhere. It has no use for consumers (because GHS classification is not meant to be understandable by ordinary people), it has no use for manufacturers/importers/etc. (because they are in fact creators of GHS data). Such data would be only for WD usage and we could tell that 'we are pro because we have GHS classification', but I really see no other applications of GHS classification except of our own satisfaction from having it. Wostr (talk) 15:19, 4 January 2018 (UTC)[reply]
    Just one word, « categories » in this terminology, seems to be subclasses (subclass of (P279) meaning) of « danger class » per http://www.unece.org/fileadmin/DAM/trans/danger/publi/ghs/ghs_rev07/French/05f_annex1.pdf . So « category 1 flammable liquid » seems to be a perfectly valid Wikidata class. We would have
    ⟨ category 1 flammable liquid ⟩ subclass of (P279)   ⟨ Flammable liquid (GHS class) ⟩
    in Wikidata. My reasoning applies at this level if not on the GHS class level. I have argumented on why it is useful on a database perspective. « however, in some cases that would not be so easy, as e.g. H271 may be present in Oxidizing liquids or in Oxidizing solids » no real problem in Wikidata whatsoever as the class hierarchy can be a directed acyclic graph (Q1195339), ie. a class can be a subclass of several others. There is no requirement whatsoever that a « message » class maps exactly one GHS class or category. « Because you have no formal rights to translate GHS classification into GHS labelling » It’s kind of a direct translation of the table linked above actually. I don’t see why we would not have a right to translate. Maybe if we do some inference trick in an infobox provide a disclaimer « inferred from the GHS category, may be overridden ». But essentially, what we should do is put a priority to datas provided with the substance in the substance item over the inferred one (that is a trick to override the automatic inference). Assuming there is few exceptions, this saves a lot of redundant datas to be written on Wikidata.
    « But GHS classification IMHO has no practical use anywhere » The number of stuffs in Wikipedia that do not have a practical use for customer is already tremendous. I fail to see how this could be an argument. There is many other kind of Wikipedians (scientists, hobbyists, curious, maybe GHS members who find more practical to use Wikidata than to dig into their own documentation if we reach a very good usability …, researcher who would like to build a live comparison of different hazard classification scheme (China vs. USA vs. EU … ). Plus this is not really more of a matter of linking a GHS category to a GHS class, for example, just a handful of statements. Not really a dispute case. author  TomT0m / talk page 16:35, 4 January 2018 (UTC)[reply]
    @TomT0m: do as you like. If you want to add GHS classification for whatever reason – sure, you just need 2–3 new properties (GHS class and/or GHS category + classification H-phrases). But I won't take part in any action that involves adding original research (and translating GHS classification into GHS labelling by any WD user is original research, because it's not as simple as you think and it's not just use some table and a bunch of exceptions; data acquired by such action would be of no value and pretending otherwise is a scam). In my home wiki I would have to block myself, if I try to add GHS labelling using classification data. I see no other way to collect GHS labelling data than having this data as 4 different properties (GHS pictogram, H/EUH-phrases, P-phrases, signal word) used as qualifiers in proposed container-like 'safety classification and labelling' property, as there is no option avaiable to Wikidata users to derive this data from GHS classification. So if you want GHS classification – sure, it's really no problem for me (as a matter of fact I've written about the possibility to add classification along with labelling data in WikiProject Chemistry), but in that case it will require 6–7 properties to store GHS data in WD and adding classification won't reduce the data that need to be written in WD. Wostr (talk) 17:38, 4 January 2018 (UTC)[reply]
    @Wostr: No it does not need a lot of properties. See Help:Classification. Actually it does not change the number of properties whatsoever. author  TomT0m / talk page 17:46, 4 January 2018 (UTC)[reply]
    @TomT0m: for God's sake, I really don't know how I can explain it in a more understandable way.
    • If you want to have GHS classification data in WD you need: GHS class/category property + H-phrases property (minimum 2 properties, if you use subclass of (P279) to model relation between class and category)
    • If you want to have GHS labelling data in WD you need: GHS pictogram + H/EUH-phrases (different from classification H-phrases) + P-phrases + signal word (4 properties; no more, no less).
    So, if you want to have both sets (labelling+classification) you need at least 6 properties. There is no option to model this data in a way that you will be able to obtain labelling data from only GHS classes and H-phrases (2 properties). Also, you cannot derive any information about classification from exisiting P728 (P728), because these phrases are not classification H-phrases but labelling H-phrases (and in labelling some phrases may be omitted, so deriving classification from this data would give you incomplete information about classification = some classes/categories would be missed in this action). If you think otherwise, then I think you have never at least read GHS and CLP and I simply don't know if continuing this discussion makes any sense. Wostr (talk) 18:07, 4 January 2018 (UTC)[reply]
    @TomT0m: okay, I think I understand what this is about. You just want to create items for classes and categories and using subclass of (P279) add them to substance items. Yep, that is possible and rather easy to achieve. I didn't think that it is all about just items, sorry. But in that case I really don't know why this discussion took place here, as this have nothing to do with proposed property and would be completely separate from GHS labelling data. Wostr (talk) 18:28, 4 January 2018 (UTC)[reply]
    Of course, H-phrases etc. could be connected in some way to classes and categories items (but I don't think that instance of (P31)/subclass of (P279) are correct here, maybe some other general property). But this whole model wouldn't give anyone information about GHS labelling (only a bunch of possibilities how such labelling might look) and it's true that such model could be some sort of representation of GHS classification scheme. I think the right place to discuss this is Wikidata:WikiProject Chemistry/Proposal:Models. Sorry, I didn't understand what this is about earlier. Wostr (talk) 18:35, 4 January 2018 (UTC)[reply]
    (edit conflict) @Wostr: Here is where we disagree. « GHS class/category property + H-phrases property (minimum 2 properties, if you use subclass of (P279) to model relation between class and category) » I need zero property. First, as « GHS category » is a subclass, adding the « GHS class » in the item is redundant as it is easy to find the class from the category. If it’s easy to find, it’s useless to add the statement in each classified substance. So we actually don’t need a « GHS class » property. I trade this with a statement on GHS classes items to state they are actually GHS classes (
    ⟨ unstable explosive ⟩ instance of (P31)   ⟨ GHS hazard class ⟩
    and on GHS categories a statement that they are GHS categories) Second, by the same reasoning that applies to use « subclass of » to model relationships between GHS classes and categories, we don’t need a specific property to class a substance on the relevant GHS category : we can use instance of (P31) or subclass of (P279). I try to argue for quite a long time that beeing generic in classification property is a good thing as there is classifications everywhere and it allows us not to bother reinventing the wheel over and over, discussing where it’s not really needed, but the idea seems hard to push :/ ) author  TomT0m / talk page 18:41, 4 January 2018 (UTC) [reply]
    For the second, the properties are already present in the current proposal, so it’s no additional property (compared to the current proposal, sorry I should have made this explicit.)
    I’m not familiar with enwiki’s policies, sorry. It’s not enwiki, it’s not even a Wikipedia here. There is objective to serve as a public data repository way beyond Wikipedias, and I always thought it’s a bad idea to import those mindset in a younger project with different goals. oops, answered
    No problem. The point about Wikidata:WikiProject Chemistry/Proposal:Models is deeply entangled with property proposals. Actually I advocate to propose whole models and not property by property, but another idea that is hard to push. author  TomT0m / talk page 18:41, 4 January 2018 (UTC)[reply]
  • I hid the discussion about GHS classification (not labelling) as I think it is not directly related to the discussed property (GHS classification may be model using subclass of (P279), but I think that GHS labelling/NFPA 704 cannot be). Also, there is Wikidata:WikiProject Chemistry/Safety classification and labelling subpage, where I placed proposed model using this property with proposed new GHS pictogram property and changes in H/P-phrases properties (string → wikibase-item datatype).
    @TomT0m: there is also GHS classification proposal based on subclass of (P279) and classes/categories items on this subpage (under option 2); I think that there are the key points of your proposals, but feel free to correct them in any way. However, I'm not sure what is your opinion about discussed property in connection with GHS labelling/NFPA 704/other labelling systems? I hope you don't mind hiding our discussion – but if you do, I'll revert this action.
    @ChristianKl: sorry for pinging, but I'm not sure whether your comments were only about using icons vs items or maybe you have some opinion about discussed property? Also, I wonder how much support have to be expressed here to be able to create property? Wostr (talk) 00:00, 8 February 2018 (UTC)[reply]
    • @Wostr: It was not only a proposal to model GHS classification, which have its interest, but also to use it to classify substances. The assumption was that it is possible it to compute automatically the messages and pictograms from the classes substances are classified into. It is way less interesting if it’s not possible to do this (I have my doubts on this, but you’re the expert and I don’t plan to really dig on the subject or code anything, so I’ll stop here :) ) author  TomT0m / talk page 08:09, 8 February 2018 (UTC)[reply]
      • I wish it could be possible for European GHS to compute labelling from the classification; it would be much easier to add just classes/categories. Maybe it is somehow possible for US GHS (which I think is more similar to original UN GHS, yet I'm not familiar with US law, so that's my guess only), but in EU GHS there is too much odd rules added to GHS (e.g. recently, while I was adding GHS labelling in pl.wiki, I found chem. compound that was formally not dangerous according to GHS criteria, had no GHS classes/categories, and yet it had GHS labelling with EUH-phrases only... that's one of a few odd things that can happen with EU GHS). Wostr (talk) 13:21, 8 February 2018 (UTC)[reply]
@TomT0m: Your proposition is focused on use of instance/subclass instead of properties to describe items but this vision just leads to the creation of dozens statements "instance of". Just take methanol: you can describe it form its structure (for example chemical compound) but also by its effects (flammable compounds, toxic compound,...) or its use (drug, fuel, solvent, disinfectant). But this is not the end when speaking about compounds classification because each system has its own definition and a flammable liquid for GHS can be not flammable for NFPA 704. So in our case you should multiply the number of possible values by the number of classification systems to be sure to represent all cases. So instead of flammable liquid you should add GHS flammable liquid and NFPA 704 flammable liquid. Just do that for classification systems like GHS (fro EU, US and other countries), NFPA 704, Hazchem, HMIS, for road transportation (ADR), for rail transportation (RID),... and I think you will understand why use of instance/subclass is just a nightmare.
That's why the use of instance/subclass classification system should be limited from my point of view especially when you are working with a general ontology and not with a specialized one. So do we agree that specific properties are more valuable than instance/subclass classification ? Snipre (talk) 14:48, 13 February 2018 (UTC)[reply]
@Snipre: « So do we agree that specific properties are more valuable than instance/subclass classification ? » this white/black opposition does not make any sense to me, the whole point is to use both the best way possible. Especially because risk classification can be performed using « subclass of » and that will help define precisely what the risk is (for example, the risk of explosion in the different national or international scheme may be defined in a different way, it’s still possible to classify then as subclass of a generic « risk of explosion ». That said, I understand your point, and in general ontologies there is properties for potentialities of objects (for OBO http://ontology.buffalo.edu/smith/articles/realizables.pdf ) and has properties such as « role » : https://www.ebi.ac.uk/ols/ontologies/ro/terms?iri=http%3A%2F%2Fpurl.obolibrary.org%2Fobo%2FBFO_0000023 and variants : https://www.ebi.ac.uk/ols/search?q=role&groupField=iri&start=0&ontology=ro .
My proposition was to use the class to infer the pictograms, as to each class maps a set of pictograms, and to use the classes as a shortcut to represent a set of pictogram and warning messages. This would not be such a nightmare in the sense that if that worked, that could spare a lot of explicit statements while retaining the possibility of putting all the information in, say, an infobox. A feature that is possible with good classes, chosen such as they allow to compress a lot of information in one statement (imagine the class of substances that are considered explosive on any considered definition). author  TomT0m / talk page 17:09, 13 February 2018 (UTC)[reply]
I experimented on that idea on Template:Implied instances that currently generates a query to find instances of a class even if the instance of (P31) statement is not put, according to statements on the class item on or its parent classes, and plan to do stuffs the other way around (if an item is an instance of a class that defines statements for its instances, consider that this item has suche a statement). author  TomT0m / talk page 17:31, 13 February 2018 (UTC)[reply]
  • I'm cautiously marking this as 'ready', as there is no votes against and the discussion whether we should add GHS classification by this property or use subclass of (P279) for this is quite beyond the scope of this discussion (in fact, this is a more general discussion whether we should use general properties like P31/P279 if it's possible or create and use many separate properties in each field) — in other words, NFPA 704, GHS labelling etc. can be added using this property (of course after some changes in existing properties and after creating some additional properties and items – which I want to do after creating proposed property), but addition of GHS classification requires further discussion in the WikiProject). Wostr (talk) 01:11, 13 March 2018 (UTC)[reply]
  • So I propose to add a constraint on the values of this new property as following:
Safety system Country Value Versions
Globally Harmonized System of Classification and Labelling of Chemicals (Q899146) EU Regulation (EC) No. 1272/2008 (Q2005334)
US Example
CN Example
UNO Example GHS 1st edition, GHS Rev.1, GHS Rev.2, GHS Rev.3, GHS Rev.4, GHS Rev.5, GHS Rev.6, GHS Rev.7,
NFPA 704 (Q208273) US 1996 version, 2007 version, 2012 version, 2017 version
UN Recommendations on the Transport of Dangerous Goods (Q7865437) UNO 1996 version, 2007 version, 2012 version, 2017 version

@ديفيد عادل وهبة خليل 2, ChristianKl, TomT0m, Wostr, Snipre:   Done: safety classification and labelling (P4952). − Pintoch (talk) 11:53, 14 March 2018 (UTC)[reply]

Retrieved from "https://www.wikidata.org/w/index.php?title=Wikidata:Property_proposal/safety_classification_and_labelling&oldid=2116575087"