Re: A Question about Agent-based Service Discovery/Selection

Web Services registry, such as UDDI, could be used to narrow down the 
search and to obtain those services that are relevant in a given industry 
category. Further filtering is also possible (via t-Model extensions to 
UDDI) to indicate that the requester prefers to obtain a list of services 
that are described in a given language (say wsdl or DAML-S). In the 
BookAirlineTicket example - a B2C scenario, a public registry of Web 
Services is conceivable for doing such pre-filtering. In a B2B setting, a 
case might be made for dealing with a (pre-selected) preferred set of 
suppliers and it is most likely more of a private registry.

Regards
Rama Akkiraju

Senior Software Engineer
Semantic e-business Middleware Group
IBM T. J. Watson Research Center
Hawthorne, NY

e-mail: akkiraju@us.ibm.com





Francis McCabe <fgm@fla.fujitsu.com>
Sent by: public-sws-ig-request@w3.org
03/05/2004 11:55 AM
 
        To:     Gregory Huczynski <greg@dcs.gla.ac.uk>
        cc:     public-sws-ig@w3.org
        Subject:        Re: A Question about Agent-based Service 
Discovery/Selection



Just a quicky response ...
You are partly right.
However, it is not monolithic, and it is also naive to assume that 
there will be *one* airline interface.

The means of interacting with a service *is* part of the semantics of 
that service. So it figures in the discovery process.

A much more interesting and pertinent issue is whether anyone is 
willing to trust such automatic discovery when spending significant 
amounts of money!

Frank McCabe

On Mar 5, 2004, at 3:00 AM, Gregory Huczynski wrote:

>
> Hi,
>
> I'm a newcomer to the area of Web Services, and I've recently started 
> investigating the current research into how services are discovered, 
> and a particular one selected that suits the user's (or application's) 
> needs. I'm interested in how these procedures can be done 
> automatically by some agent-like software entity.
>
> I've read quite a few position-style papers (e.g. Semantic Web 
> Services, McIlraith et al, IEEE Intelligent Systems, 2001), which 
> advocate describing and advertising services using semantic richer 
> descriptions written in DAML+OIL and OWL. The argument is given that 
> these richer descriptions will be more amenable to automated 
> manipulation and reasoning than those currently provided by discovery 
> mechanisms such as UDDI.
>
> However, from reading these papers, I am slightly confused by a basic 
> issue. Let's consider an example given in McIlraith et al's paper: .. 
> "A user might say, for example, "Find a service that sells airline 
> tickets between San Francisco and Toronto and that accepts payment by 
> Diner's Club credit card"". I assume that with the 
> agent-OWL-description approach, this problem would be solved by an 
> agent reasoning over the semantic markup of the available services, 
> identifying those that meet the user's constraints.
>
> However, in order for the agent to book a ticket automatically from a 
> service, it must be able to communicate with it: it must know and 
> understand the service's interface (specified, say, in WSDL). So, my 
> question is this - for problems like the example above, before any 
> reasoning can be done to ensure we meet user constraints, is the 
> service search-space first narrowed down to only those services that 
> implement a WSDL (communications) interface understood by the agent?
>
> For example, in solving the problem above:
>
> - Assume that a standard BookAirlineTicket WSDL interface has been 
> defined. Lots of airline ticket services export this interface.
> - Assume that the user's agent knows and understands the 
> BookAirlineTicket interface (it could communicate with a 
> BookAirlineTicket service)
> - In solving the problem, the agent first narrows its search to those 
> advertised services that export the BookAirlineTicket interface
> - The agent then reasons over the identified BookAirlineTicket 
> services' descriptions, and identifies those that meet the user's 
> constraints.
> - The agent then offers these matched services to the user, or books 
> the ticket itself.
>
> Basically, in order for a service to be discovered, selected and used 
> automatically, do we not require an initial search-space constraining 
> step based on the notion of known / previously-defined service 
> function interfaces?
>
> Apologies for writing such a long email for such a basic question!
>
> Thanks in advance
>
> Gregory Huczynski
> Department of Computing Science
> University of Glasgow
>