Raphaël Troncy

• Semantic Web and Multimedia Technologies (multimedia metadata standards)
• Knowledge Representation (languages, descrition logics, conceptual graphs)
• Ontology Modeling (methodologies, tools)
• Ontology Alignment

K-Space

• Kick-off Meeting - 11-13 January 2006, London (UK)
• Second General Meeting - 27-29 March 2006, Amsterdam (Netherlands)
• Technical Meeting - 27-28 April 2006, Athens (Greece)
• Third General Meeting - 28-30 August 2006, Graz (Austria)
• Fourth General Meeting - 4-5 December 2006, Athens (Greece)
• Technical Meeting - 8-9 February 2007, Koblenz (Germany)
• Review Meeting - 8-9 March 2007, London (UK)
• Fifth General Meeting - 12-13 March 2007, Koblenz (Germany)
• Sixth General Meeting - 11-12 June 2007, Prague (Czech Republic)

NewsML and the Semantic Web

More and more news are produced and consumed each day. News are still mainly textual stories, but they are more and more often illustrated with graphics, images and videos. News can be further processed by professional (newspapers), directly accessible for web users through news agencies, or automatically aggregated on the web, generally by search engine portal and not without copyright problems. For easing the exchange of news, the International Press Telecommunication Council (IPTC) is currently developping the NewsML2 Architecture whose goal is to provide a single generic model for exchanging all kinds of newsworthy information, thus providing a framework for a future family of IPTC news exchange standards [1]. This family includes NewsML, SportsML, EventsML, ProgramGuideML or a future WeatherML. All are XML-based languages used for describing not only the news content (traditional metadata), but also their management, packaging, or related to the exchange itself (transportation, routing).

However, despite this general framework, some interoperability problems occur. News are about the world, so their metadata might use specific controlled vocabularies. For example, IPTC itself is developping the IPTC News Codes [2] that currently contain 28 sets of controlled terms. These terms will be the values of the metadata in the NewsML2 Architecture. But the news descriptions often refer to other thesaurus and controlled vocabularies, that might come from the industry, and all are represented using different formats. From the media point of view, the pictures taken by the journalist come with their EXIF metadata [3]. Some videos might be described using the EBU format [4] or even with MPEG-7 [5].

Durint the latest general IPTC meting, CWI has presented some Semantic Web technologies to solve this interoperablity problems. More precisely, we have developped an OWL ontology capturing the semantics of the NewsML specification and a converter that automatically transform feeds in NewsML into RDF. We have also made some converters for transforming the IPTC News Codes into various SKOS thesaurus. We are currently closely collaborating with AFP in order to bring NewsML on the Semantic Web for better retrieving, presenting and processing news on the Web.

References

[1] News Architecture (NAR) for G2-standards, http://www.iptc.org/NAR/
[2] The IPTC NewsCodes - Metadata taxonomies for the news industry, http://www.iptc.org/NewsCodes/
[3] EXIF: Exchangeable Image File Format, Japan Electronic Industry Development Association (JEIDA). Specifications version 2.2 available in HTML and PDF
[4] EBU: European Broadcasting Union, http://www.ebu.ch/
[5] MPEG-7: Multimedia Content Description Interface, Standard No. ISO/IEC n°15938, 2001.

Relevant publications

• Newsletter of the Industry Board of the EU FP6 Knowledge Web NoE, month November

National Projects

• OPALES

CNRS "Specific Action" Working Groups

• Semantic modeling and multimedia indexing (2002-2003)
• Semantic Web (2001-2002)

The manipulation of audio-visual documents is both a difficult and a specific task that requires to deal with intermediary representations. The temporal nature of audio-visual force to use a description to enrich the documents with metadata and then exploit them. The digitalizing of audio-visual documents allows new use of the content such as "intelligent" search, dynamic recomposition, or personalized access. The technical system that produces these services has to be integrated and must link the content with its description.

We defend that a representation of both the structure and the content of the documents is necessary. By structure, we mean the documentary structure, that is the mereological organization of the elements that compose the document, whereas the content refers to a conceptual structure, that is a characterization of these elements. This double representation highlights the need for a homogeneous description format that is usable, very expressive and efficient for the machine. After a review of the current propositions for modeling audio-visual documents, coming from the document engineering side and the knowledge representation side, we demonstrate that none of the studied languages can deal with these two aspects in a satisfactory way. We then propose a general architecture that allows to describe formally the structure and the semantics of audio-visual documents and that will generate a knowledge base on which we will perform some reasoning.

This architecture is composed of an audio-visual ontology, whose a part is translated into a documentary language in order to control the logical structure of the document, and a domain ontology for describing the semantics of the content. Consequently, two ontologies have been built: the generic audio-visual ontology and a cycling ontology which is the application domain of our architecture. We have developed the DOE tool (Differential Ontology Editor) that implements the chosen methodology for building the ontologies. Finally, we demonstrate the relevance of our approach with two experiments on an annotated corpus of video and for which an implementation of knowledge bases is proposed, showing hence the kind of inference we can make.

Actuellement, le Web contient d'importantes quantités d'informations couvrant tous les sujets imaginables. Le problème qui était avant de savoir si une information, même très spécifique, était disponible sur le Web, est maintenant devenu comment retrouver cette information. Apporter du sens intelligible et exploitable par des machines aux documents devrait leur permettre d'utiliser l'information présente, d'améliorer les techniques de recherche, et donc de faire du Web une gigantesque base de connaissance. Les langages de représentation de connaissance sont de bons candidats si l'on souhaite décrire le contenu de documents. L'action Escrire a d'ailleurs pour objectif d'en comparer trois. Parmi eux, la représentation de connaissances à objets apparaît particulièrement adapté lorsqu'il s'agit de représenter des connaissances complexes sur un domaine en cours d'étude. On pourra alors manipuler plus efficacement une base de documents en les indexant par leur contenu (ou leur sens). Les documents pertinents seront ramenés à partir de requêtes structurées tirant parti du formalisme de représentation de connaissance (hiérarchie de classes, mécanismes de classification...).

Nous avons d'abord observé le lien étroit existant entre la nature de la connaissance à représenter et le type du document. Nous avons aussi pu voir que plus que le contenu, c'est l'application résultante qui va décider des éléments à représenter. Nous avons donc essayé d'imaginer quels types de requêtes un utilisateur est susceptible de poser, ce qui nous a conduit à proposer un langage de requêtes. Un corpus de travail a été utilisé pour mettre en oeuvre les choix effectués. Il concerne les interactions géniques chez la drosophile pendant son processus de segmentation. Le système de représentation de connaissances à objets Troeps gère les connaissances contenues dans les documents. Un évaluateur de requêtes a été intégré à ce système pour permettre de l'interroger.