Skip to Main content Skip to Navigation
Theses

Un modèle d'indexation relationnel pour les graphes conceptuels fondé sur une interprétation logique

Abstract : The idea of establishing relations between objects and representing them in the knowledge base of an information system is a feature of all approaches in Artificial Intelligence. However, the majority of knowledge representation formalisms do not fully exploit the semantic richness of these relations and their associated behavior. In information retrieval, the processing of these relations is not at all better performed and the impact of their taking into account at the time of the matching phase has never been established though it is true that many approaches take into account their presence in the document and tend to represent them as part of the indexing process. Nevertheless, retrieval of structured and complex documents makes it necessary, more than ever, to use the relation semantics and properties, next to a solid and expressive indexation language. By use of a study of the new demands to which information retrieval has to answer nowadays, we propose a relational indexing model for documents. The approach consists of considering that an indexing term is based on complex concepts where semantic connectors are considered to be operators, or relations that allow to build new expressions representing new concepts or situations. The proposed model is not limited to relation representation, but also offers a general framework that mentions the general principle for the handling of these relations and the taking into account of their properties in a retrieval process based on a logical approach. The model proposed here is made up of two components: the representation language allowing for a relational indexing approach, and inference rules that make it possible to direct the correspondence process by use of the language. We use the situation theory as representation language and a relevance derivation system which is based on the axiomatisation of the notion of correspondence between documents and the query to take into account the relations. An interesting feature of this model is that it leads to extensions of some knowledge representation formalisms, with useful notions in information retrieval. The limitations of the terminological logics, otherwise used as the formal basis of the proposed relational indexing approach, can be thus overcome. At the same time, the complexity of the treatments associated to this family of logics does not allow to use them as an operational model. Consequently, we propose the conceptual graph formalism as a good compromise between the complexity of theorem provers and the simplicity of algebraic approaches. This formalism is thus considered, by means of an adequate logical interpretation, as being an implementation of an extended terminological logic and of the indexing model. Our approach was implemented on a conceptual graph handling platform realised on top of the OO-DBMS O2. The resulting prototype RELIEF was tested on an image collection. The experimentation proved the practicability and well-foundation of the approach.
Document type :
Theses
Complete list of metadatas

Cited literature [262 references]  Display  Hide  Download

https://tel.archives-ouvertes.fr/tel-00004902
Contributor : Thèses Imag <>
Submitted on : Thursday, February 19, 2004 - 3:54:38 PM
Last modification on : Friday, November 6, 2020 - 4:04:32 AM
Long-term archiving on: : Friday, April 2, 2010 - 8:35:57 PM

Identifiers

  • HAL Id : tel-00004902, version 1

Collections

UJF | CNRS | IMAG | UGA

Citation

Iadh Ounis. Un modèle d'indexation relationnel pour les graphes conceptuels fondé sur une interprétation logique. Autre [cs.OH]. Université Joseph-Fourier - Grenoble I, 1998. Français. ⟨tel-00004902⟩

Share

Metrics

Record views

346

Files downloads

626