[
Lists Home |
Date Index |
Thread Index
]
John Cowan wrote:
>
>...
>
> Think about some more loosely coupled applications than remote procedure
> call: data mining, for example. In data mining, the application that
> reads through the data does not care what the intention was of the
> process that wrote it, which is probably long since dead.
I think that the data mining application does care partially. For
instance sometimes my customers have data that was thirty or forty years
old and the semantics of the tags have shifted over time. In order to
reliably process that data we had to know what the tags meant "back
then". They don't care what process the data was involved in, or what
the goals of the day were, or who created the data or why. But they care
*what was meant* by <xaaa>5</xaaa>. It is very important whether it
meant "height in feet" or "weight in stones".
Schemas (of both the syntactic and semantic variety) can slow the
semantic drift.
> ...
> Now extend this notion to real time, where you have processes that are
> steadily churning out knowledge to consumers that can do whatever they
> want with it. For example, Reuters Health narrowcasts news articles each
> day to its customers, but does not care what they do with them: the
> articles can be displayed as-is on a customer Web site, incorporated into
> a database, or printed out for use by a human editor.
I claim that they still care *what you meant* by the data. And further,
if they can get you to standardize your name->meaning mapping with that
of other data providers, they will save a bundle of money which enhances
the value of your service to them (modulo hundreds of political issues).
--
Paul Prescod
|
