Skip to Main content Skip to Navigation

Représentations de connaissance: de l'approximation à la confrontation

Jérôme Euzenat 1
1 SHERPA - Object Knowledge Bases
Inria Grenoble - Rhône-Alpes
Abstract : A knowledge representation formalism aims at modelling a precise domain. There are several such formalisms and, within one of them, there can be numerous models of the same domain. The present report is devoted to the study of the relationships between multiple representations of the same situation. It describes the work of the author between 1992 and 1998 progressing from the approximation relation to the confrontation of the representations. First, the notion of an approximation between object-based representations is developed mainly with regard to the taxonomy-based mechanisms (classification, categorisation, clustering). A classification scheme allows to consider these mechanisms in a homogeneous way and puts forward the approximation relation between the initial representation and the final one. It is shown how to build a knowledge representation system according to the principles of classification schemes. The second chapter introduces the advantages and problems of providing alternative taxonomies (organising the same set of objects) in a knowledge representation system. It considers some choices that can be made in implementing multiple taxonomies. These taxonomies are rephrased in terms of classification schemes. Granularity is the topic of the third chapter. Contrary to the other chapters, the representation formalism is an algebra of binary relations enabling to express the relations (in time or space for instance) between objects. The relationship between representations is constrained since the representations are ordered by the level of detail they consider. The rules governing such representations are presented. Last, the fourth chapter aims at confronting the various representations in order to take advantage of them (i.e. building a consistent and consensus representation). The goal of the presented works is the implementation of a system for co-operatively building consensus knowledge bases in which the users build a common knowledge base from the content of their individual ones. Two particular problems are considered: the design of a revision mechanism for object-based knowledge representation formalisms allowing the users to deal with inconsistency and the design of a knowledge submission protocol ensuring the gathering of consensus knowledge. Although limited, this insight into possible studies of the relationships between representations reveals the non-imperative character of the developed solutions, which are well suited to the interaction between the user and the representation system.
Document type :
Habilitation à diriger des recherches
Complete list of metadatas

Cited literature [27 references]  Display  Hide  Download

https://tel.archives-ouvertes.fr/tel-00340958
Contributor : Jérôme Euzenat <>
Submitted on : Monday, November 24, 2008 - 10:27:30 AM
Last modification on : Thursday, March 5, 2020 - 4:49:39 PM
Document(s) archivé(s) le : Monday, June 7, 2010 - 8:23:58 PM

Identifiers

  • HAL Id : tel-00340958, version 1

Collections

Citation

Jérôme Euzenat. Représentations de connaissance: de l'approximation à la confrontation. Intelligence artificielle [cs.AI]. Université Joseph-Fourier - Grenoble I, 1999. ⟨tel-00340958⟩

Share

Metrics

Record views

726

Files downloads

672