Web-Scale Workflow
88 Published by the IEEE Computer Society 1089-7801/09/$25.00 © 2009 IEEE IEEE INTERNET COMPUTING
Editor : M . Br ian Blake • mb7@george town .edu
T he paradigm known as service-oriented computing supports developing Web-based applications (or Web services) that mirror
the capabilities that human actors traditional-
ly perform.1 Service mashups (also referred to
as Web or enterprise mashups2,3) currently re-
ceive considerable attention from the service-
oriented computing community. We can create
a service mashup by simultaneously executing
two or more services, thus combining the result-
ing messages into an integrated, coherent view
of multiple data objects. For example, we can
combine Web services from an organization that
provides a mapping capability with capabilities
from a parcel delivery service. Both services’ in-
tegrated data provisions can illustrate the path
of specific parcels that, for example, get lost in
transit (see Figure 1).
Current attention in this area aims to create
tools and techniques that instrument the mash-
up process (that is, if a user has identified two or
more complementary services, then these tools
will facilitate the integration of their data pro-
visions). Subsequently, these tools visualize the
results via a common image.4 Perhaps a more
interesting capability, and one that might yield
nondeterministic results, is a new approach that
could recommend potentially complementary
groups of services, on demand. Such “service
mashup recommendations” let us discover new
knowledge in much the same vein as knowledge
discovery in databases (KDD) but without some
of its more intractable challenges. This knowl-
edge discovery in services (KDS) approach might
produce benefits similar to KDD, but it also in-
troduces new issues.
Predicting Service Mashups
Data integration in general is difficult5 when
we consider that openly available data might
have significant instances of syntactic and se-
mantic mismatch. However, Web services are
defined via Web Service Description Language
(WSDL)-based documents in which service pro-
visions are decomposed into lower-level WSDL
parts (WSDL parts are the same as parameters
and returns in traditional software methods).
As such, KDS benefits from these more struc-
tured specifications.6,7 Furthermore, some Web
service specifications identify these parts via
mark-up tags, whereas others define them se-
mantically (as with the Web Ontology Language
for Services [OWL-S; www.daml.org/owl-s/] and
WSDL-S [www.w3.org/Submission/WSDL-S/]).
Consequently, recommending complementary
services is, in itself, less constrained than the
requirements associated with inferring real
equivalences across disparate data. The remain-
ing question is “How do you define groups of
services that might be complementary and eligi-
ble for mashup?” In earlier work,8 my colleagues
and I defined service mashup candidates in mul-
tiple ways. Corresponding with Figure 2,
A• represents a group of Web services that
have one or more related output parts;
B• represents a group of Web services that
have any combination of equivalent parts,
Knowledge Discovery
in Services

M. Brian Blake • University of Notre Dame
Service mashups can be useful in understanding Web-scale workflows. Although
creating a service mashup shares similar challenges with data integration, a
more exciting aspect of this area is the ability to predict which services are
viable candidates for a mashup. Such “service mashup recommendations” can
enable knowledge discovery, an approach the author calls knowledge discovery
in services (KDS).