The goal of the semantic web is to take advantage of formalised knowledge (in languages like RDF) at the scale of the worldwide web. In particular, it is based on ontologies which define concepts used for representing knowledge on the web, e.g., for annotating a picture, specifying a web service interface or expressing the relation between two persons.
However, it is likely that different information sources will use different ontologies. It is thus necessary to find correspondences between concepts of these ontologies. The operation of finding correspondences is called ontology matching and its result is a set of correspondences called alignment [1]. Alignments are used for importing data from one ontology to another or for translating queries.
In general, a correspondence is a triple (e1,e2,r) such that e1 is an entity, e.g., a class, a property, an instance, of a first ontology, e2 is an entity of the second ontology and r expresses the relation between e1 and e2. This relation may be the equivalence or inclusion between these entities or more complex relations, e.g., that one entity contains another one.
Alignments obey a precise semantics that is naturally extended to networks of aligned ontologies, defining what can be deduced from a particular set of aligned ontologies [2].
Matching ontologies is very difficult. Hence, existing alignments are worth reusing. For that purpose, we have developed alignment servers for sharing alignments on the web [3]. They can generate alignments on demand but can also store alignments that have been created manually or certified as correct. In order to take advantage of a potentially wide set of available alignments and ontologies, it is necessary to consider operations for alignment management [4].
In this doctoral work, we propose to study different mechanisms that help manipulating alignments: composition, revision and negociation.
Composition is a way to deduce new alignments from existing ones. If there exists an alignment between ontology O1 and ontology O2, and another one between O2 and a third ontology O3, we want to decide which correspondences hold between O1 and O3. The operation that returns this set of correspondences is called composition. For instance, consider the correspondence 1:MappleLeaf subClassOf 2:Leaf between ontology O1 and ontology O2, as well as the correspondence 2:Leaf partOf 3:Tree between O2 and O3. Composing these two correspondences could yield 1:MappleLeaf partOf 3:Tree between O1 and O3. Since relations between ontology entities may be arbitrarily complex, composition obeys particular rules that must be investigated. One way to model this is by using algebras of relations [5].
Alignment composition can thus be reduced to combining correspondences with regard to their relations and the structure of related entities. Other operations related to composition must also be considered: conjunction is used when the same correspondence is obtained by different means and disjunction is used when several correspondences are competing. These operations must be designed as a coherent system.
Alignment revision can be considered when the introduction of an alignment in a network of aligned ontologies leads to inconsistency. In such a case, consistency can be restored by modifying the alignment or the remainder of the network of aligned ontologies. A similar operation called update is considered when ontologies in the network evolve. Instead of computing new alignments from scratch, it is worth trying to update the previously existing alignments.
We want to consider revision and update approaches in the style of [6] or [7]. Some preliminary work have been done on similar approaches [8,9], but we want to reach a deaper understanding of revision and update in networks of aligned ontologies. In particular, the operations may be different if an ontology is modified or if an alignment is modified, in such cases the impact of the modification may target differently alignments and ontologies. One direct outcome of the work on update and composition is the matching of ontology versions [10].
Finally, argumentation frameworks define arguments for or against propositions, that can themselves be arguments. These arguments are evaluated in order to assess the acceptable sets of propositions which are supported by a solid set of arguments. We participed in applying this work to ontology alignments [11], where agents negotiate the alignments between their ontologies before communicating: they issue arguments in favor or against the correspondences contained in an alignment and the system decide what is the most convenient set of correspondences. We want to consider more thoroughly the connection between arguments and their (indirect) consequences on the resulting network semantics. For instance, one can take a set of arguments into account only if they preserve the consistency of the considered networks of aligned ontologies.
The successful candidate will study these possible operations and their behaviour. Their respective benefits and drawbacks will be investigated. In particular, these operations are expected to be consistent with the alignment semantics [2]. The result of this work could be implemented in our alignment representation system [12] which grounds our Alignment server.
Références:
[1] Jérôme Euzenat, Pavel Shvaiko, Ontology
matching, Springer-Verlag, Heildelberg (DE), 2007
[2] Antoine Zimmermann, Jérôme Euzenat, Three semantics for distributed systems and their relations with alignment composition, in: Proc. 5th conference on International semantic web conference (ISWC), Athens (GA US), Lecture notes in computer science 4273, pp16-29, 2006
[3] Jérôme Euzenat, Alignment infrastructure for ontology mediation and other applications, in: Martin Hepp, Axel Polleres, Frank van Harmelen, Michael Genesereth (eds), Proc. 1st ICSOC international workshop on Mediation in semantic web services, Amsterdam (NL), pp81-95, 2005
[4] Jérôme Euzenat, Adrian Mocan, François Scharffe, Ontology alignments: an ontology management perspective, in: Martin Hepp, Pieter De Leenheer, Aldo De Moor, York Sure (eds), Ontology management: semantic web, semantic web services, and business applications, Springer, New-York (NY US), pp177-206, 2008
[5] Jérôme Euzenat, Algebras of ontology alignment relations, Proc. 7th ISWC, Karlsruhe (DE), Lecture notes in computer science 5318:387-402, 2008
[6] Carlos Alchourrón, Peter Gärdenfors, David Makinson, On the logic of theory change: partial meet contraction and revision functions, Journal of symbolic logic 50(2):510-530, 1985
[7] Hirofumi Katsuno, Alberto Mendelzon, On the difference between
updating a knowledge base and revising it, Proc. 2nd international conference on principles of knowledge representation and reasoning, Cambridge (MA US), 1991
[8] Özgür Özçep, Towards principles for ontology integration,
Proc. 5th FOIS, pp137-150, 2008
[9] Maciej Zurawski, Alan Smaill, Dave Robertson, Bounded ontological
consistency for scalable dynamic knowledge infrastructures, Proc. 3rd ASWC, Bangkok (TH), Lecture Notes in Computer Science 5367:212-226, 2008
[10] Natalya Noy and Mark Musen. PromptDiff: A fixed-point algorithm
for comparing ontology versions. In Proc. 18th National Conference
on Artificial Intelligence (AAAI), pages 744-750, Edmonton (CA),
2002.
[11] Loredana Laera, Ian Blacoe, Valentina Tamma, Terry Payne, Jérôme
Euzenat, Trevor Bench-Capon, Argumentation over ontology
correspondences in MAS, Proc. 6th International conference on
Autonomous Agents and Multiagent Systems (AAMAS), Honolulu (HA US),
pp1285-1292, 2007
[12] Jérôme Euzenat, An API for ontology alignment,
in: Proc. 3rd conference on international semantic web conference
(ISWC), Hiroshima (JP), Lecture notes in computer science
3298, pp698-712, 2004
Qualification: Master or equivalent in computer science.
Researched skills:
Environnement: The doctoral work is to be carried out in the Exmo team, a reputed team within the semantic web research area and particularly on ontology matching. We are in contact with the most important teams in France and Europe on these topics and this work could be the occasion of cooperation on sharing alignments.
Doctoral school: Doctoral school MSTII, Grenoble.
Advisor: Jérôme Euzenat (Jerome:Euzenat#inrialpes:fr)
Group: Exmo, INRIA Grenoble Rhône-Alpes
Hiring date: September 2009.
Place of work: The position is located at INRIA Rhône-Alpes, Montbonnot (near Grenoble, France) a main computer science research lab, in a stimulating research environment. Research will be carried out in the Exmo team under the supervision of Jérôme Euzenat. It will require the involvement of the candidate in related projects.
Duration: 36 months
Salary: 1537 EUR/month net (including full health insurance and social benefits) upgraded to 1620 EUR/month the 3rd year.
Contact: For further information, contact Jerome:Euzenat#inrialpes:fr.
Procedure:
Visit
INRIA's
presentation (including FAQ and forms) .
File: Provide Vitæ, motivation letter and references.
Please be sure to apply on INRIA online application system