Skip to Main content Skip to Navigation
Theses

Modélisation de la pertinence en recherche d'information : modèle conceptuel, formalisation et application

Abstract : Information retrieval systems aim at allowing a user to reach documents that contribute to solve the information problem that prompted the use of the system. The system can be viewed as a predictive apparatus, that predicts the relevance of documents for the user. To do this, systems generally rely on pieces of evidence that are of a topical nature, namely the set of keywords that can be obtained from the query entered by the user. The system implements thus a matching function that maps documents onto queries, and that models the topical dimension of relevance. Although, uses ans users of information retrieval systems are getting more varied, as well as information media that is not exclusively textual anymore. We draw two consequences from these evolutions. Firstly, the assumption that the topical relevance factor is paramount and as such, the only factor actually modelled in systems, does not hold anymore. Many other relevance factors are involved in information retrieval, that greatly influence systems performance in the context of a real-life use of a system. These factors strongly depend on the individual who performs the search, and on the situation in which the individual stands, and this fact requestions the traditional modelling of relevance as a normalizing topical matching between documents and queries. Secondly, the interactive nature of information retrieval system use, contributes to the definition of the user's situation, and as such, participates in the system's overall performance. A number of interactive features in systems are directly concerned with the modelling of relevance and with IR-specific matters -- viz. the number of retrieved documents that the user faces, the precision or exhaustivity of the result, etc. This thesis relies on user relevance studies to define a model for the design of system relevance that accounts for those relevance factors involved in the interactive use of systems, and for the need for adapting the matching function to the particular retrieval situation at hand. Hence, we define three user-oriented functional duties for information retrieval systems: allow for the detection of relevant documents retrieved, allow for the understanding and comprehending of the reasons for documents to be retrieved, and allow for the production of useful feedback for the iterative improvement of the query. The concept of relevance schema substitutes to the traditional query, to play the role of interfaces between the system relevance and the user. This relevance schema is comprised of two kinds of parameters that allow tuning the system in relation with the retrieval situation encountered: on the one hand, semantic parameters encompass the topical dimension together with other types of relevance criteria, and on the other hand, pragmatic parameters account for relevance factors related to the conditions in which the user achieves the interactive tasks required. We apply this design to an image retrieval application, with a multi-faceted indexing available that allow for the definition of relevance criteria that go beyond topical relevance criteria. We show via our prototype how the system can adapt to situations occurring during a retrieval session.
Document type :
Theses
Complete list of metadatas

Cited literature [54 references]  Display  Hide  Download

https://tel.archives-ouvertes.fr/tel-00004938
Contributor : Thèses Imag <>
Submitted on : Friday, February 20, 2004 - 3:54:34 PM
Last modification on : Friday, November 6, 2020 - 4:04:57 AM
Long-term archiving on: : Friday, September 14, 2012 - 10:20:46 AM

Identifiers

  • HAL Id : tel-00004938, version 1

Collections

UJF | CNRS | IMAG | UGA

Citation

Nathalie Denos. Modélisation de la pertinence en recherche d'information : modèle conceptuel, formalisation et application. Interface homme-machine [cs.HC]. Université Joseph-Fourier - Grenoble I, 1997. Français. ⟨tel-00004938⟩

Share

Metrics

Record views

748

Files downloads

458