[Data-modeling] Olympics domain for review

[Robert Cook]

On Mar 26, 2008, at 9:43 PM, Kurt Bollacker wrote:
>
> On Wed, Mar 26, 2008 at 05:41:37PM -0700, Robert Cook wrote:
>> I believe we should use abstract types minimally, and instead use
>> domain-specific models.  Ultimately, to support the queries you
>> suggest, we could have meta-structures in the schema that equate
>> properties among different types that are semantically equivalent.
>>
>> Abstract types are a kind of contingency bet where you are assuming
>> you will need generality and you are willing to take a near-term hit
>> in complication and user confusion.  I think the near-term cost of
>> that bet is too high in most cases.  If Freebase gains enough
>> information where cross-domain queries start to make sense, then the
>> higher-order semantics can be added without impinging on the freebase
>> users who care about their specific domain.
>>
>> Over-abstraction is like premature optimization.  It's nice to think
>> you "got it right" from the beginning, but that almost never happens.
>
> Abstraction is a one way function.  We can always easily build more
> general properties out of more specific ones.  It's very hard to make
> more specific ones from more general.  From a graphd performance
> perspective it's not a disaster in a MQL query to combine all "events"
> for each sport into a query to find "all the events Michael Jordan has
> won". To pick apart the event types from the more general query is
> much more complex to parse in the general case.
>
> 								kurt :-)	

Yes, this is a much clearer statement of what I was trying to say.

R