Skip to Main content Skip to Navigation
New interface

Vers un outil sémantique d’autocodage qualitatif pour l’évaluation de l’acceptabilité des innovations

Abstract : The techniques of Natural Language Processing (NLP) and the methods of qualitative analysis of textual data share epistemological similarities. However, qualitative analysis does not fully benefit from the potential of NLP. In particular, works aiming at an actual automation of qualitative data coding remain quite rare. This thesis aims at investigating the potential of different NLP techniques for this purpose and for a task that requires a certain degree of human expertise. It aims at creating a tool that can be used in an industrial context and for a specific analysis method that allows to evaluate the acceptability of innovations. This method uses a codebook composed of 20 codes that have a higher semantic complexity than those traditionally used in computer-aided qualitative analysis.We explore ways to achieve this task through a bottom-up and then a top-down approach. For the former, we perform a lexicometric exploration on a corpus of old study data to define the lexical profile of the expected data for each code. Then, we treat the problem as a classification task by testing statistical classifiers of several types. We also investigate the possibilities offered by the projection of a syntaxico-semantic resource on the corpus.We then follow a less conventional top-down approach. For this one, we carry out an expert-informed modelling, in the form of an ontology, of the acceptability evaluation qualitative interview paradigm. This modelling is matched with a lexicon built in an ad hoc way. We thus propose an original architecture for a semantic analysis tool in which the ontology's triples are used as a support for textual interpretation and which subsets constitute classification rules. We achieve building a hyperspecialized analysis tool which performances exceed those obtained by machine learning on our training corpus. This tool is carried to the point of operationalization, by being integrated into an autocoding platform, for the purpose of setting up a continuous learning process.
Document type :
Complete list of metadata
Contributor : ABES STAR :  Contact
Submitted on : Tuesday, June 7, 2022 - 2:55:24 PM
Last modification on : Thursday, June 9, 2022 - 3:37:59 AM
Long-term archiving on: : Thursday, September 8, 2022 - 6:54:25 PM


Version validated by the jury (STAR)


  • HAL Id : tel-03689721, version 1



Doriane Simonnet. Vers un outil sémantique d’autocodage qualitatif pour l’évaluation de l’acceptabilité des innovations. Linguistique. Université Grenoble Alpes [2020-..], 2022. Français. ⟨NNT : 2022GRALL002⟩. ⟨tel-03689721⟩



Record views


Files downloads