Définition et évaluation de modèles d'agrégation pour l'estimation de la pertinence multidimensionnelle en recherche d'information

Abstract : The main research topic of this document revolve around the information retrieval (IR) field. Traditional IR models rank documents by computing single scores separately with respect to one single objective criterion. Recently, an increasing number of IR studies has triggered a resurgence of interest in redefining the algorithmic estimation of relevance, which implies a shift from topical to multidimensional relevance assessment. In our work, we specifically address the multidimensional relevance assessment and evaluation problems. To tackle this challenge, state-of-the-art approaches are often based on linear combination mechanisms. However, However, these methods rely on the unrealistic additivity hypothesis and independence of the relevance dimensions, which makes it unsuitable in many real situations where criteria are correlated. Other techniques from the machine learning area have also been proposed. The latter learn a model from example inputs and generalize it to combine the different criteria. Nonetheless, these methods tend to offer only limited insight on how to consider the importance and the interaction between the criteria. In addition to the parameters sensitivity used within these algorithms, it is quite difficult to understand why a criteria is more preferred over another one. To address this problem, we proposed a model based on a multi-criteria aggregation operator that is able to overcome the problem of additivity. Our model is based on a fuzzy measure that offer semantic interpretations of the correlations and interactions between the criteria. We have adapted this model to the multidimensional relevance estimation in two scenarii: (i) a tweet search task and (ii) two personalized IR settings. The second line of research focuses on the integration of the temporal factor in the aggregation process, in order to consider the changes of document collections over time. To do so, we have proposed a time-aware IR model for combining the temporal relavance criterion with the topical relevance one. Then, we performed a time series analysis to identify the temporal query nature, and we proposed an evaluation framework within a time-aware IR setting.
Document type :
Theses
Complete list of metadatas

Cited literature [245 references]  Display  Hide  Download

https://hal-univ-tlse2.archives-ouvertes.fr/tel-01249652
Contributor : Bilel Moulahi <>
Submitted on : Saturday, January 2, 2016 - 1:53:09 PM
Last modification on : Thursday, October 24, 2019 - 2:44:06 PM
Long-term archiving on : Friday, April 15, 2016 - 3:32:48 PM

Licence


Distributed under a Creative Commons Attribution - NonCommercial - NoDerivatives 4.0 International License

Identifiers

  • HAL Id : tel-01249652, version 1

Collections

Citation

Bilel Moulahi. Définition et évaluation de modèles d'agrégation pour l'estimation de la pertinence multidimensionnelle en recherche d'information. Recherche d'information [cs.IR]. Université Toulouse III Paul Sabatier, 2015. Français. ⟨tel-01249652⟩

Share

Metrics

Record views

431

Files downloads

3680