Composing Web Services on the Basis of Natural Language Requests
Alessio Bosca, Andrea Ferrato, Fulvio
Corno
Politecnico di Torino
Torino, Italy
{alessio.bosca, fulvio.corno}@polito.it
Ilenia Congiu, Giuseppe Valetto
Telecom Italia Lab
Torino, Italy
{ilenia.congiu,giusepp.valetto}@tilab.com
Abstract
The introduction of the Semantic Web paradigm in
Service-oriented Architectures enables explicit
representation and reasoning about services, via a
semantically rich description of their operations. We
propose an approach towards service selection and
composition based upon the interpretation of user
requests expressed through an informal human-
computer interaction interface that employs
(restricted) Natural Language.
1. Introduction
This paper focuses primarily on a synergy that we
recognized between Natural Language Processing
(NLP) techniques and Semantic Web Services, and on
how such synergy can empower an automated, on-
demand service composition, starting from informal
user requests expressed in Natural Language.
We employ OWL-S [1] annotations to provide a
formal representation of service and operation
semantics and an ontological classification of the Web
Services in our portfolio. These annotations form a
semantic vocabulary that is also exploited for the
interpretation of user requests, in order to derive
intentions, and then extract the major functional
requirements and constraints about the required service
composition.
Our approach follows from two main assumptions:
that user requests are relative simple, in structure and
terminology, to be expressed with a controlled subset
of natural language and that a common ontological
vocabulary can be established, and is consistently
applied to all entries in the Service Catalog.
2. A semantic service Catalog in OWL-S
Our Service Catalog has as its foundation a common
ontological framework expressed in OWL-S, which
provides us with a high-level formalism to
semantically describe and categorize each atomic
service operation.
We have come up with a way to model Effects in
OWL-S in order to enable the selection of service
operations that satisfy some user requests or needs. Our
modeling approach promotes the description of Effects
in terms of the computing task performed by each
OWL-S AtomicProcess (AP) element that models the
operations exposed by services in our Catalog; when
we interpret a request, we provide for the dynamic
composition of those AP by generating a direct acyclic
graph, where such AP act as nodes and the relations
among their IOPEs (Inputs, Outputs, Preconditions and
Effects) establish arcs. At invocation time, we translate
that abstract graph into an executable service, using
OWL-S grounding to bind each Atomic Process to a
concrete operation. To this end, we have implemented
an ad hoc ontology called Effects (see bottom of
Figure 1).
Additionally, we focused on I/O parameters
semantic, referring to a set of concepts collected in
another ad-hoc ontology called IOtypes. In order to
reason on inputs and outputs for the automatic,
semantic-based composition of operations we extended
the OWL-S model with a couple of bi-directional
properties that allow us to link processes to their I/O
parameters and parameter to processes that can
produce or consume them.
Figure 1 - Atomic Process: findCinema
Proceedings of the IEEE International Conference on Web Services (ICWS’05)
0-7695-2409-5/05 $20.00 IEEE