Re: How will the semantic web emerge


I'm not sure we're not talking about different applications (and perhaps 
a stricter definition of "taxonomy" than I was using).  Certainly all 
the machine-learning approaches you describe are in practical (and 
increasing) use today, but it seems to me that "most approaches based on 
static, manually maintained taxonomy, are dead" goes awfully far.  This 
is particularly true given your mention of Oracle, SQL Server, and DB2, 
all of which are based on the use of static, manually maintained 
relational schemas.  While these are not "taxonomies" in the sense that 
they define subsumption and other such relationships, they certainly 
constitute contracts on the meaning of various pieces of data for the 
purpose of interoperability (and "vocabularies" in the RDF sense).  I 
don't expect the use of such contracts (or other pre-agreements on what 
terms mean) to go away very soon, if at all.  However, that doesn't mean 
they can't be used in conjunction with machine-learning.  It seems to 
me, for example, that a useful "hybrid" approach would be to apply 
machine-learning techniques to allow such schemas to evolve from their 
initial definitions with use, hence becoming less static and manually 


Joshua Allen wrote:
>>>I think all of this is way too pie in the sky for the semantic web.
> Yes
>>I tend to agree with your point here that a lot of what will go on the
> To be specific, I was talking about the sorts of real-world applications
> we have today, for example those machine-learning approaches which ship
> in the box with Oracle, SQL Server, and DB2.
> A specific example is the "people who bought items in your cart also
> liked..."  Or sites which dynamically select ads based on the types of
> content you view or publish.  Sites which recommend ringtones based on
> analysis of aggregate behavior of millions of users.
> SQL Server shipped for a number of versions with a platform for doing
> taxonomy-based English query.  This never really caught broad adoption,
> and other vendors found the same.  OTOH, when we started shipping
> machine-learning based platform pieces: Bayesian classifiers,
> clustering, etc. we found that these were quickly adopted in many
> real-world situations.  Google's pagerank is another example of machine
> learning applied to real-world problem (collaborative filtering).  You
> can find many approached based on machine learning in practical use.
> Most approaches based on static, manually maintained taxonomy, are dead.

Received on Monday, 19 December 2005 14:48:17 UTC