Skip to Main content Skip to Navigation
Theses

Segmentation thématique de texte linéaire et non-supervisée :
Détection active et passive des frontières thématiques en Français

Abstract : This research belongs to the Natural Language Processing (NLP) field and more specifically focuses on topic text segmentation. The originality of this thesis consists in integrating to an unsupervised topic text segmentation method syntactic, semantic and stylistic information. This work present a linear approach of topic text segmentation based on a vectorial representation of the sentence coming from a deep morpho-syntactic and semantic analysis. This representation is then used to compute distance between potential topic segment while integrating stylistic information. During this research an application has been developed, this application allows users to test the approach various parameters, but also some others methods that have been tested during this research. Our model has been evaluated using an automatic evaluation and a manual evaluation. Our manual evaluation leas us to develop a specific evaluation protocol for the task based on precise parameters. In both automatic and manual evaluation our results as good and sometimes even better than some of the most popular algorithms.
Document type :
Theses
Complete list of metadatas

Cited literature [69 references]  Display  Hide  Download

https://tel.archives-ouvertes.fr/tel-00364848
Contributor : Alexandre Labadié <>
Submitted on : Friday, February 27, 2009 - 2:43:45 PM
Last modification on : Monday, October 19, 2020 - 11:12:01 AM
Long-term archiving on: : Tuesday, June 8, 2010 - 9:12:07 PM

Identifiers

  • HAL Id : tel-00364848, version 1

Collections

Citation

Alexandre Labadié. Segmentation thématique de texte linéaire et non-supervisée :
Détection active et passive des frontières thématiques en Français. Autre [cs.OH]. Université Montpellier II - Sciences et Techniques du Languedoc, 2008. Français. ⟨tel-00364848⟩

Share

Metrics

Record views

463

Files downloads

3043