[Developers] mediated properties
Robert Cook
robert at metaweb.com
Thu Jul 19 21:59:49 UTC 2007
Augusto --
You are asking a very good question. A "mediator" is a data pattern
that is represented as a Compound Value Type in Metaweb. In other
words, mediators are CVTs that act like an "annotated relationship"
between two types. One example is "Marriage", which joins two people
together that also stores add start and end dates as metadata.
Currently, we don't have a way to annotate which of the two
properties represent the "relationship" between two types. That is,
a mediator is more of a data idiom that makes sense to the modeler,
but has no formal representation in the data store.
We've received feedback similar to yours in the past, and we're
currently working on a way of making that distinction (while
simultaneously solving a few other problems.) We don't yet have an
ETA, but know that it's a real priority for us.
Robert
On Jul 19, 2007, at 2:37 PM, Augusto Callejas wrote:
> hi-
>
> i ran that query but all properties have value "true"
> for the "disambiguator" property hint.
>
> from the type documentation:
>
> Musical Group Membership is a compound value type which defines the
> relationship between a Musical Artist (a band or collaboration)
> and its
> Musical Group Members (musicians).
>
> is there a difference when you say it mediates for all of the
> properties
> versus when the documentation says it defines relationships between
> musical artist and group member? i'm looking for how that
> relationship
> is captured in the type information. since the disambigurator
> property
> hint is the same for all properties, it doesn't define the
> relationship
> i'm looking for.
>
> thanks,
> augusto.
>
>
> On 7/19/07 10:52 AM, "Darin Wilson" <darin at metaweb.com> wrote:
>
>> Hi Augusto-
>>
>> A mediating type, or complex value type (CVT), actually mediates for
>> all of the properties that the type contains; however, the type
>> designer can choose which of those properties should be displayed
>> when an instance of the type linked from another topic. If you look
>> at the type definition of /music/group_membership, you'll see that
>> each of the properties has the "Display as disambiguator" box
>> checked. This means that the property will be displayed, if the data
>> is available.
>>
>> If you look at the instance list for /music/group_membership, you'll
>> see that some of the instances have years and other don't. That's
>> because the "Period (start)" and/or "Period (end)" properties haven't
>> been filled out yet. If the data were available, it would be
>> displayed.
>>
>> http://www.freebase.com/view/filter?id=%2Fmusic%2Fgroup_membership
>>
>> To see this programmatically, you can use the following query. The /
>> freebase/type_hints/mediator property on the type will tell you if
>> it's a CVT or not. The /freebase/property_hints/disambiguator on the
>> properties will tell you if the value of the property should be
>> displayed.
>>
>> {
>> "query":[{
>> "id":"/music/group_membership",
>> "type":"/type/type",
>> "/freebase/type_hints/mediator":null,
>> "properties":[{
>> "id":null,
>> "/freebase/property_hints/disambiguator":null
>> }]
>> }]
>> }
>>
>> HTH,
>> Darin
>>
>>
>> On Jul 19, 2007, at 10:05 AM, Augusto Callejas wrote:
>>
>>> hi-
>>>
>>> regarding mediating types, how do i determine what two other types
>>> it is mediating for? for instance, there is the complex type:
>>>
>>> http://freebase.com/view/filter?id=/music/group_membership
>>>
>>> reading the type description reveals that it relates these two
>>> types:
>>>
>>> /music/group_member (via property /music/group_membership/member)
>>> /music/musical_group (via property /music/group_membership/group)
>>>
>>> how do i determine this programmatically? at first, i was going to
>>> use that fact that these two properties have "unique":true, but
>>> so is the case for two other properties:
>>>
>>> /music/group_membership/start
>>> /music/group_membership/end
>>>
>>> how can i make the distinction?
>>>
>>> thanks,
>>> augusto.
>>>
>>>
>>> On 6/28/07 3:46 PM, "Dae Park" <daepark at metaweb.com> wrote:
>>>
>>>> We perform a "schema query" for each type a topic is an instance
>>>> which includes a query for each property and their
>>>> expected_type. If
>>>> the expected_type is marked as a mediator then we go another ply
>>>> out
>>>> and ask for the property's expected_type's disambiguating
>>>> properties.
>>>>
>>>> {
>>>> id: #topic_id
>>>> type: [{
>>>> id: null,
>>>> properties: [{
>>>> id: null,
>>>> master_property: null,
>>>> reverse_property: null,
>>>> unique: null,
>>>> expected_type: {
>>>> id: null,
>>>> "/freebase/type_hints/mediator": null,
>>>> properties: [{
>>>> master_property: null,
>>>> reverse_property: null,
>>>> id: null,
>>>> unique: null,
>>>> "/freebase/type_hints/disambiguator": true
>>>> }}
>>>> }
>>>> }]
>>>> }]
>>>> }
>>>>
>>>> When displaying these mediator properties (sub_properties) from a
>>>> topic's property:
>>>>
>>>> if the topic_property and sub_property are reciprocal of one
>>>> another:
>>>> if sub_property is unique:
>>>> skip
>>>> else:
>>>> foreach value in sub-property:
>>>> if value == topic we're viewing:
>>>> continue:
>>>> display value
>>>>
>>>>
>>>> Two properties (prop1, prop2) are reciprocal's of one another if:
>>>>
>>>> prop1.reverse_property == prop2
>>>> or
>>>> prop2.master_property == prop1
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>> On Jun 28, 2007, at 1:15 PM, Steve Sak wrote:
>>>>
>>>>>
>>>>> How does the freebase.com "person" web pages "know" to display the
>>>>> name
>>>>> of the spouse/sibling or the name of the organization a person
>>>>> worked
>>>>> for under "employment history"? Hard-coded?
>>>>>
>>>>> Nick Thompson wrote:
>>>>>> in general, i don't think your question has a well-defined
>>>>>> answer -
>>>>>> there's no reason for the system to believe that
>>>>>> /business/employment_tenure/company is more relevant than
>>>>>> /business/employment_tenure/title, for example. you could be
>>>>>> looking
>>>>>> for a list of people at a particular company, in which case
>>>>>> the job
>>>>>> title would more interesting.
>>>>>>
>>>>>> so, you might rule out /business/employment_tenure/person from
>>>>>> context
>>>>>> as chris described, but that doesn't determine a unique "other
>>>>>> side".
>>>>>> that's the nature of compound value types - if a relationship
>>>>>> only
>>>>>> had two sides, it would probably be modeled as a single link
>>>>>> instead
>>>>>> of being mediated by a compound value object.
>>>>>>
>>>>>> nick
>>>>>>
>>>>>> Steve Sak wrote:
>>>>>>> So, if I'm looking at /people/person/employment_history how
>>>>>>> would I
>>>>>>> determine that the relevant property is
>>>>>>> /business/employment_tenure/company?
>>>>>>>
>>>>>>>
>>>>>>> Alec Flett wrote:
>>>>>>>> I want to add that checking by is only valid if you know you're
>>>>>>>> coming
>>>>>>>> in and out on the same property (as is the case with marriage)
>>>>>>>> - if
>>>>>>>> the properties are different, you need to look at master/
>>>>>>>> reverse
>>>>>>>> properties because you can imagine a director who also
>>>>>>>> produced a
>>>>>>>> movie - if you were just going by ids, looking at the
>>>>>>>> relationship of
>>>>>>>>
>>>>>>>> "Who is the /director/ of this movie /produced by/ Martin
>>>>>>>> Scorsese?" -
>>>>>>>> you might get Martin Scorsese back, but that doesn't mean that
>>>>>>>> "director" is the reverse of "produced by".
>>>>>>>>
>>>>>>>> Alec
>>>>>>>>
>>>>>>>> Christopher R. Maden wrote:
>>>>>>>>> Steve Sak wrote:
>>>>>>>>>
>>>>>>>>>> To restate the question: given a mediator object how do I
>>>>>>>>>> determine
>>>>>>>>>> which property is the one it's mediating? For example,
>>>>>>>>>> type A
>>>>>>>>>> has a
>>>>>>>>>> mediator M that has a bunch of properties mediates between A
>>>>>>>>>> and
>>>>>>>>>> another
>>>>>>>>>> object. How do I determine which property of M is the one
>>>>>>>>>> that
>>>>>>>>>> points
>>>>>>>>>> to the other object being mediated?
>>>>>>>>>>
>>>>>>>>> Mediators, or compound value types, are not especially
>>>>>>>>> different
>>>>>>>>> from any other type; the UI treats them a little differently,
>>>>>>>>> but
>>>>>>>>> that’s it.
>>>>>>>>>
>>>>>>>>> But I think you are asking about instances rather than types,
>>>>>>>>> correct? In other words, Arnold Schwarzenegger connects to a
>>>>>>>>> marriage, and that marriage connects to two people, both
>>>>>>>>> Arnold
>>>>>>>>> Schwarzenegger and Maria Shriver. How can you tell which
>>>>>>>>> of the
>>>>>>>>> spouses is the one you started from?
>>>>>>>>>
>>>>>>>>> The only way to do that, really, is by looking at the IDs.
>>>>>>>>>
>>>>>>>>> {
>>>>>>>>> "query": {
>>>>>>>>> "id": null,
>>>>>>>>> "name": "Arnold Schwarzenegger",
>>>>>>>>> "spouse_s": [
>>>>>>>>> {
>>>>>>>>> "spouse": [
>>>>>>>>> {
>>>>>>>>> "id":null,
>>>>>>>>> "name":null
>>>>>>>>> }
>>>>>>>>> ],
>>>>>>>>> "type":"/people/marriage"
>>>>>>>>> }
>>>>>>>>> ],
>>>>>>>>> "type":"/people/person"
>>>>>>>>> }
>>>>>>>>> }
>>>>>>>>>
>>>>>>>>> {
>>>>>>>>> "result": {
>>>>>>>>> "spouse_s": [
>>>>>>>>> {
>>>>>>>>> "spouse": [
>>>>>>>>> {
>>>>>>>>> "id": "#9202a8c04000641f80000000001da07e",
>>>>>>>>> "name": "Maria Shriver"
>>>>>>>>> },
>>>>>>>>> {
>>>>>>>>> "id": "#9202a8c04000641f8000000000006567",
>>>>>>>>> "name": "Arnold Schwarzenegger"
>>>>>>>>> }
>>>>>>>>> ],
>>>>>>>>> "type": "/people/marriage"
>>>>>>>>> }
>>>>>>>>> ],
>>>>>>>>> "type": "/people/person",
>>>>>>>>> "id": "#9202a8c04000641f8000000000006567",
>>>>>>>>> "name": "Arnold Schwarzenegger"
>>>>>>>>> }
>>>>>>>>> }
>>>>>>>>>
>>>>>>>>> You can see that one of the spouses in the marriage has the
>>>>>>>>> same ID
>>>>>>>>> as the person with whom you started, and that the other is
>>>>>>>>> different.
>>>>>>>>>
>>>>>>>>> Unfortunately, there’s not really an automatic way to say
>>>>>>>>> “get all
>>>>>>>>> of the values of these properties except the ones that point
>>>>>>>>> back
>>>>>>>>> to the parent within the scope of this query.”
>>>>>>>>>
>>>>>>>>> HTH,
>>>>>>>>> Chris
>>>>>>>>>
>>>>>>>
>>>>>
>>>>> --
>>>>> Steven G Sak
>>>>> Applied Minds, Inc.
>>>>> 11718 Bowman Green Drive
>>>>> Reston, Virginia 20190
>>>>> o: (703) 483-2207
>>>>> c: (703) 626-2557
>>>>>
>>>>> _______________________________________________
>>>>> Developers mailing list
>>>>> Developers at freebase.com
>>>>> http://lists.freebase.com/mailman/listinfo/developers
>>>>
>>>> _______________________________________________
>>>> Developers mailing list
>>>> Developers at freebase.com
>>>> http://lists.freebase.com/mailman/listinfo/developers
>>>
>>>
>>> _______________________________________________
>>> Developers mailing list
>>> Developers at freebase.com
>>> http://lists.freebase.com/mailman/listinfo/developers
>>
>> _______________________________________________
>> Developers mailing list
>> Developers at freebase.com
>> http://lists.freebase.com/mailman/listinfo/developers
>
>
> _______________________________________________
> Developers mailing list
> Developers at freebase.com
> http://lists.freebase.com/mailman/listinfo/developers
More information about the Developers
mailing list