Skip to Main content Skip to Navigation

Construction semi-automatique de ressources pour la fouille d'opinion

Résumé : Identifying satisfaction triggers among customers is a crucial need in today's business world, as a strong customer relationship is now a most vital asset. The domain of opinion mining, in which this thesis falls into, offers several methods to answer this need. These methods, however, require a continuous update of specialized resources which are the cornerstone of many opinion mining tools. The objective of this work is to develop acquisition and structuration strategies for these resources, which can be lexicons, morphosyntactic rules or annotated data. Each of these items presents its own extraction difficulties, on top of the general issue of their update in a language- or domain-specific setting. Indeed, language constraints are fundamental in opinion mining, so the proposed methods must take these into account. First, we study the core elements from which opinion expressions are built in customer feedback. This study leads us to suggest a new modelisation of opinion mining as a sequence labeling task. We then compare the benefits of each type of resource through a benchmark of several opinion mining methods, and conclude that the best performing strategy is a hybrid approach. Finally, we present results for resource acquisition methods that answer not only the needs of opinion mining but also the constraints from the industrial setting in which this work has been conducted.
Complete list of metadatas

Cited literature [161 references]  Display  Hide  Download
Contributor : Joseph Lark <>
Submitted on : Tuesday, November 7, 2017 - 7:08:00 PM
Last modification on : Wednesday, June 24, 2020 - 4:19:36 PM
Document(s) archivé(s) le : Thursday, February 8, 2018 - 3:11:50 PM


Files produced by the author(s)


  • HAL Id : tel-01630619, version 1


Joseph Lark. Construction semi-automatique de ressources pour la fouille d'opinion. Traitement du texte et du document. Université de Nantes, Faculté des sciences et des techniques., 2017. Français. ⟨tel-01630619⟩



Record views


Files downloads