W3C home > Mailing lists > Public > public-sws-ig@w3.org > December 2003

Re: UDDI and semantics: CMU OWL-S/UDDI Mapping

From: Massimo Paolucci <paolucci@cs.cmu.edu>
Date: Mon, 01 Dec 2003 18:58:29 -0500
Message-ID: <3FCBD5A5.2070306@cs.cmu.edu>
To: www-sws-ig <public-sws-ig@w3.org>

Evan,

I heard the critique that you point out many, many times.  You are actually right in saying that we match only the inputs and outputs and that may not be enough.  The way I see it is the following:

1. minimally, we have to add matching of preconditions and effects. Unfortunately, when we developed our algorithm, early 2001, there was no clear way in which to express the types of predicates that we would like to express using DAML.  There were proposals (RuleML being one) that provided such a language, but they were not on the way to become standards and they were not endorsed by the DAML-S coalition.  Now a rule language is coming out, and we should be able to push the matching beyod inputs and outputs.

2. There is a deeper problem with matching for capabilities.  As I see it there are two ways to express capabilities: one is explicit by saying that a service S is a ComputerSeller,  the other is implicit by saying that it is a transformation from a state to another.  Ideally, you would like to have ontologies that support the explicit representation, but the problem is that those ontologies are hard to come by.  This is because 1. there are an awful amount of functions out there, 2. you cannot really define functions without defining the things that they modify,  so you actually need two ontologies one of things, the other of functions.  To the extent there there are ontologies out there, they represent things, not functions.   What we are trying to do at CMU is to push the implicit way to represent functions so that there is no need for ontologies of functions.  I do not know how far can we go, and I am afraid that even adding preconditions and effects will not be enough, but it is worth the attempt.

3. the discovery process is more complicated than a simple match.  There may be many web services providing the function that you want, and you have to select among them and find a way to use them.  Hopefully, if there is a false positive selected by the matchmaker, it will be weeded out by the interaction process.

I hope that helped.

Cheers,

--- Massimo




Evan K. Wallace wrote:

>Paul Denning quoted Massimo Paolucci:
>
>  
>
>>>The result is a Web service registry that acts as UDDI but it can also 
>>>answer queries such as "which web service can do X for me?",  which 
>>>cannnot really answered by the current UDDI.
>>>      
>>>
>
>and went on to question this claim regarding UDDI.  I have also read
>[1] and do not believe that the paper demonstrated that CMU's DAML-S 
>enhanced webservice matchmaking system actually answers the question "which 
>web service can do X for me?"  Rather it answers the question, "Are the inputs
>and outputs (parameters) of this service compatible with the inputs 
>and outputs of the service I want?"   For example the definition for the 
>advertised service could have matched anything that took in a price and 
>returned a car.  A car rental service would have matched the request, when 
>what was actually wanted was a car selling service.  Many other types of 
>services with much more divergent functions could also have matched the 
>underspecified request.  
>
>I agree that given sufficiently rich definitions of the parameters this
>blackbox description approach could give some confidence that  
>an advertised service will be something like the requested service.  But 
>no detailed parameter ontology was presented which could have enabled this 
>and it could be pretty difficult to create one using only RDF(S) and OWL. 
>
>I would be interested in hearing from Massimo or someone else who worked
>on the system about how they plan to refine this method to reduce the
>false positives.  Do you plan to define deeper parameter ontologies or are 
>you waiting for an accepted language for specifying preconditions and 
>effects to address the issue?
>
>-Evan
>
>Evan K. Wallace
>Manufacturing System Integration Division
>NIST
>ewallace@nist.gov
>
>
>[1] Massimo Paolucci, Takahiro Kawamura, Terry R. Payne, Katia 
>Sycara; "Semantic Matching of Web Services Capabilities." 
><http://www-2.cs.cmu.edu/%7Esoftagents/papers/ISWC2002.pdf> In 
>/Proceedings of the 1st International Semantic Web Conference/ (ISWC2002)
>
>
>  
>
Received on Monday, 1 December 2003 19:01:35 GMT

This archive was generated by hypermail 2.2.0+W3C-0.50 : Sunday, 16 March 2008 00:10:53 GMT