Français Anglais
Accueil Annuaire Plan du site
Home > Research results > Dissertations & habilitations
Research results
Ph.D de

Ph.D
Group : Artificial Intelligence and Inference Systems

Mapping discovery in a semantic peer to peer system: application to SomeRDFS

Starts on 01/09/2006
Advisor : REYNAUD, Chantal

Funding : BDI CNRS
Affiliation : Université Paris-Saclay
Laboratory : LRI IASI

Defended on 17/09/2010, committee :
BERND AMANN, Professeur au LIP6 (examinateur)
ZOHRA BELLAHSENE, Professeur au LIRMM (examinateur)
SALIMA BENBERNOU, Professeur au LIPADE (rapporteur)
CHRISTINE FROIDEVAUX, Professeur au LRI (examinateur)
OLLIVIER HAEMMERLÉ, Professeur à l'IRIT (rapporteur)
CHANTAL REYNAUD, Professeur au LRI (directrice de thèse)

Research activities :

Abstract :
The richness of answers to queries posed to PDMS depends on the number of mappings
between ontologies of different peers. Increasing this number can improve responses
to queries. This is the problem considered in this thesis.

We aim at discovering semantic links between ontologies of different peers. This
problem, known as ontology alignment, is specific in peer-to-peer systems in which
ontologies are not completely known a priori, the number of ontologies to align is
very large and alignment should be done without any centralized control.

We propose semi-automatic techniques for identifying: (1) mapping shortcuts corresponding
to a composition of existing mappings and (2) new mappings which cannot be inferred in
the current state of the system. These techniques are based on the use of reasoning
mechanisms of PDMS and filtering criteria restricting the number of pairs of elements
to align.

Mappings shortcuts are identified from the analysis of the trace of queries asked by
users, but also after application of criteria considering their usefulness. The discovery
of new mappings consists of identifying the elements of the ontology of a given peer
that are judged interesting and then of selecting the elements from distant peers with
which it is relevant to align with. The proposed alignment techniques are either
adaptations of existing technology or innovative techniques exploiting the specificities
of our framework.