[Tagging] General tagging system problems (was: shop=confectionery / pastry / candy / sweets)

Daniel Koć daniel at xn--ko-wla.pl
Tue May 12 22:32:54 UTC 2015


W dniu 12.05.2015 21:50, Michał Brzozowski napisał(a):

> 1) Don't reinvent the wheel. See how "competitors" have tackled a
> problem (Gonna elaborate very widely on that when I'll have enough
> examples and time to write).

I don't know the competition and I'm curious how do they deal with it. 
After all it's not that easy, because we try to cover the whole planet 
and it's a very diverse place.

Examples would be great!

> 2) Tag to users' expectations, not to your definitions. But the
> consequence would be a substantial reduction in the mailing list
> traffic :-P

I guess what we're trying to do on this list is defining things 
according to multiple layman points of view.

However you're right: usually from some point discussion just gets 
deeper and deeper into horrible details because it's easier to invent 
another "cases" than try to keep things clean, general and 
user-friendly.

> 3) Ontology should be simple and rather general. Being too particular
> while incomplete is a plague of current shop and services tagging
> system.

Not only incomplete, but also overlapping in many cases - which is even 
worse, because you can always fill the missing place, but it's much 
harder to redefine already used tagging schemes and narrow down 
definitions (BTW: that seems to be exactly the problem with 
confectionery etc.).

***

I had the same intuition lately and that's what I said on the Talk list 
about how should good tagging system look like:

1. It should be more uniform (like "amenity=school" -> "landuse=school"
for the school areas).

2. It should be more cascading/hierarchical (like in
"construction=highway + highway=service + service=parking_aisle").

3. It should be more granular (no more
"amenity=green_poodle_with_6_legs", just because it's a very common
case! Rather "amenity=poodle + colour=green + legs=6").

4. It should allow mixing different forms and functions (like in
"building=church + amenity=place_of_worship", because they can be
disconnected, like "building=church + tourism=museum").

5. It should treat parallel types of objects as first class citizens
(kind of "amenity=police + amenity=school" for police academy should be
possible, since this amenity is equally a teaching place _and_ a police
place - the same for multiple names: we can make it "name=A;B" if really
needed, but the semicolon is our last resort and there's no consensus if
we should use numbering schemes like "name1=A + name2=B" or "name:1=A +
name:2=B" instead).

[ https://lists.openstreetmap.org/pipermail/talk/2015-March/072349.html 
]

In my opinion current tagging system fails especially at 3. - it is not 
granular enough and I see no way of repairing it directly. So I invented 
the idea of smaller, more universal "bricks" like:

- area
- food
- drink
- sweet
- sleep
- children
- education
- service
- religion
- building
- shop
- railway
- bus
- vehicle
- home
- art

...and so on. Remember, this is just an illustration of the problem, not 
the final list! This new "vocabulary" should be created by carefully 
analyzing, generalizing and extracting from current system to 
re-implement the knowledge we use now. Then we should be able to 
construct things like:

children + education + building (= school building)
vehicle + education (= driving school)
area + tree (= forest/wood)
building + sleep (= hotel/hostel/...)

and many more much easier, avoiding too much overlapping and letting 
things be general when needed or when the mapper can not be sure. If we 
made also the ontology tree for this vocabulary (like "children is kind 
of person" and "bus is a vehicle"), we could have sane, granular and 
extendable system. Wiki would no longer be the fat, necessary phone book 
like it's now, because it would be easier to remember the system and to 
extend them according to the rules.

To make the transition as smooth as possible, we could establish that 
some new combinations are "reserved" for old objects (like "children + 
education" is exactly "amenity=school").

Of course it's all just general sketch to be refined and examined. I 
wrote more about this idea in this post:

https://lists.openstreetmap.org/pipermail/talk/2015-March/072375.html

I know it looks like a big task, but I see no shorter way to achieve 
better coherency and usability of our tagging system in the long run.

-- 
"The train is always on time / The trick is to be ready to put your bags 
down" [A. Cohen]



More information about the Tagging mailing list