Re: Action 10 - What Is A POI? - "Place-Oriented Information"?

On Wed, Nov 10, 2010 at 10:11 AM,  <gary.gale@nokia.com> wrote:
> Based on what's been discussed on the public mailing list, I've drawn
> together a definition and description of what constitutes a POI. This will
> no doubt be cause for much discussion and debate but we need a good starting
> point to drive and frame the discussion ...
> Best
> G
>
> What Is A POI?
> Wikipedia defines a POI as a Point Of Interest ... a "specific point
> location that someone may find useful or interesting". But for the purposes
> of this Working Group, we need a more subtle and complex definition.
> A POI is part of a loosely coupled and inter-related geographical terms,
> comprised of (in generalised order of scope and granularity) Locations, POIs
> and Places.
> Location
> A Location is a geographical construct; a physical fixed point on the
> surface of the Earth. It could also be used to describe a fixed point on the
> surface of another celestial body but for the purposes of this Working
> Group, we'll restrict the scope to terrestrial geographies. A Location is
> described by a centroid (a longitude and latitude in a widely adopted
> system, such as WGS-84) and an extent, either a Minimum Bounding Rectangle
> or a vector set. A Location is temporally persistent, it does not generally
> change over time.
> POI
> A POI is a human construct, describing what can be found at a Location. As
> such a POI typically has a fine level of spatial granularity. A POI has the
> following attributes ...
> 1. A name
> 2. A current Location (see the commentary below on the loose coupling of POI
> and Location)
> 3. A category and/or type
> 4. A unique identifier
> 5. A URI
> 6. An address
> 7. Contact information
> A POI has a loose coupling with a Location; in other words, a POI can move.

I like the idea of breaking out Location, and Place, and these kinds
of fields seem the right kind of thing. But I'm not yet comfortable
with POI itself. It's a slippery notion!

Perhaps the key distinction here isn't quite between 'geographical'
and 'human' constructs, but between terms that directly name aspects
of the world, versus terms that name kinds of information about that
world. The former might range from very geographical, objective,
physical things to more human constructs such as neighbourhood. The
latter makes explicit a level of indirection, and allows the
representation to be talked about explicitly. I think this is why the
concept of POI is somehow slippery when we try to pin it down.

For me, POI is much more in the latter case. "Count questions" (How
many Xs...?) can help flesh out the difference.

We can sensibly ask:

* How many streets, churches, fire hydrants, mountain tops, traffic
blockages, classical music concerts on next saturday, vegan
restaurants; canals or houseboats are there in [some defined notion
of] Amsterdam right now?

Each of these definitions is slippery in a different way, and
different agencies, groups etc might define them differently. Yet the
questions remain primarily about the world, albeit expressed using
imperfect, debatable terminology that might need clarifying.

If we assume some working consensus of specified definitions (active
churches in the x,y and z faiths; fire hydrants serviced by the civic
authority and known to be recently tested; traffic blockages reported
in the last hour and believed to be still affecting drivers but not
bikes;  etc etc.), each of these questions has factual answers. Now we
would get different answers depending on who we ask, which database we
query, how much money or time we spend asking, what our policy is
towards risk and noise in the data etc., or the exact notion we're
querying for. But the basic scenario is factual questions about the
world, answered in loose or precise form depending on context. Note
that as we get more precise ("reported in the last hour (and not
reported as fixed subsequently)"), characteristics of information and
communication start to sneak into the scenario. This is a good thing -
it means we have useful work to do!

If we ask instead:

* how many POIs are there in [some defined notion of] Amsterdam right now

I don't believe that really has a direct factual answer, without one
crucial qualifier: we need to say which collection of information
we're talking about. How many traffic blockage POIs came back in our
last database lookup? How many Fire Hydrant POIs were described in the
appendix to the 2010 hydrant QA report? How many upcoming classical
concert POIs were attached to that that last email newsletter I
received, or embedded in the concert hall's iCalendar or RSS/Atom
feed? How many POIs were stored on the DVD I've just bought entitled
'Mountaintops of the Western Netherlands?". That's a different
numerical question to the question of how many mountains are there in
the Netherlands, although the answers are likely to be related. And in
this last case, zero-ish.

The same worldly questions and themes crop up in both stories, but
when we talk about POIs we're emphasising the information about the
world as an artifact of direct interest, and in our case technical
standardisation; rather than a transparent means-to-an-end, where the
end is 'information about the world'.

By making this indirection explicit, that POIs are informational
entities, I think this eases one of our biggest conceptual problems:
how we deal with different levels of detail. From the example on
weds's call talking about a building, and Gary's desk in the building,
and even some item on that desk of Gary's in that room in that
building in that street. And for the AR guys, for professional GIS,
architecture and city planners alike, these distinctions matter. These
are all identifiable worldly entities, potentially of interest,
potentially described in a variety of standard computer formats. It
makes sense to ask factual questions like 'how many rooms in the
building', but not 'how many POIs'; we can ask 'how many POIs
describing things in that room are there in this particular dataset?'
or 'give me POIs at the granularity of DesktopObject for this area'.

This is all a longwinded way of suggesting that "POIs" are better
thought of as aspects of the *map* rather than the *territory*.
However the usual expansion of POI as "Point of Interest" hides this,
and makes us think of POIs as objective characteristics of the world
around us, countable, comparable, etc. without being set in the
context of some description, dataset or map.

If we think of POI as "Place-Oriented Information" it makes their
information-dependency much more explicit. I suspect this will help us
think through mashup-oriented issues like "ok, we have one restaurant
but 5 POIs in our system that relate to it; one's a photo, two are
reviews, one comes from a health inspector's report and another is a
3d building plan". The "POIs" (also pieces of information...) all
relate to the same spatial zone, but they carve it up quite
differently; some treat it (the photos) as an area that reflects
light, some as a service or organization/business that can be
reviewed, paid money, sued, and one as a building occupying physical
space (perhaps with others also inside it).

We want a POI standard that allows all these kinds of information
about "the restaurant" to be brought together to serve end-user
scenarios, and to make life easier on the technologists who'll
facilitate this. But we also don't want our POI standard to be
fiendishly rich, modelling fine-grained distinctions explicitly such
as "business" versus "building" within the W3C spec. I expect to see
systems that do draw those distinctions to be able to answer "how many
businesses?" "how many buildings?", "how many businesses in this
building?" kinds of question. I hope they'll be able to answer them in
part from indexing W3C POI descriptions and other extension data or
linked files (CityGML etc). But the more I think about it, the more I
reckon we should reserve POI as a technical term for talking about
those underlying data items used to answer questions, display maps and
AR views and so on, rather than talk as if POIs are actually out there
in the world. The actual points of interest are of course out there in
the world; pieces of place-oriented information live in our computers,
phones, files and data networks.

cheers,

Dan

Received on Friday, 19 November 2010 09:44:18 UTC