UMR CNRS 7253

* Language: English

* Author: Marco Luca Sbodio

* Title: Planning Web Agents: Combining Intelligent Agents and Semantic Web Languages for Service Planning

* Defense: October 2009

* Abstract:

This thesis explores the use of standard Semantic Web Languages to fully and compactly characterize services, thus enabling Intelligent Agents to select and combine them in plans. We propose SPARQL (the standard Semantic Web query language) as a formal language to describe the preconditions and effects of services, as well as the goals of agents. We show that SPARQL query evaluation can be used to check the truth of preconditions in a given context, construct the effects that will result from the execution of a service in a context, and determine whether a service execution with those results will satisfy the goal of an agent. The contributions of this thesis are as follows:

• The hybrid modal logic GpDL combining the dynamic component of traditional Propositional Dynamic Logic with SPARQL graph patters. GpDL-formulae can represent services, agents goals, and are used to model the planning problem. We ground the possible worlds semantics of GpDL-formulae into the SPARQL computational model.
• The embedding of SPARQL expressions defining services preconditions and effects into existing SemanticWeb services formalisms (OWL-S and SAWSDL), and RESTful Web services.
• The architectural and implementation description of a services index, which, inspired by traditional Web index, harvests services descriptions from the Web, builds internal data structures with services representational models, and support automatic services discovery by answering agents queries. The services index exploits the SPARQL expressions defining the services preconditions and effects to build its data structures, and answers SPARQL-based queries.
• The architectural and implementation description of a SPARQLent, a goal directed planning agent. The functioning of a SPARQLent is based on the computational model of GpDL; its goal is a SPARQL graph pattern, and its incomplete representation of the world state is an RDF graph. A SPARQLent interacts with a services index during planning, it can reason about services preconditions (graph patters relaxation), and compose services sequences through regression planning with cost-based heuristics.
• The validation and evaluation of the proposed approach in three different domains. First, a real world use case in the e-Government domain, where we use a services index and a SPARQLent to solve the problem of selecting assistance and welfare services for citizens, thus showing that our approach is general enough to deal also with generic services. Second, an experiment in the domain of collaboration services, where we use a SPARQLent to support the creation of dynamic and personalized collaborative environments. Finally, the evaluation of our approach with respect to the currently available OWL-S Services Retrieval Test Collection, which is intended to support the evaluation of OWL-S services matchmaking algorithms.