Skip to Main content Skip to Navigation
Master Thesis

Stratégies d'enrichissement sémantique d'analyse d'opinions : contributions logicielles et linguistiques

Abstract : Since 2019 Eloquant has been working on new opinion extraction tool, combining machine learning and linguistic rules. As the first version were inconsistent and opinions, often badly identified, the Sémantique team considered using complementary labelling strategies.
We have developed a second version of the opinion analysis project, using an approach that reconciles machine learning and the integration of semantic annotations. First, we have reconstructed the training corpora, in order to make them more balanced. Then, the analysis of the errors of the previous version of the project, we adapted the symbolic rules and created new ones to make the tool as many effective as possible. Finally, the evaluation of the new version of the opinion analysis project allowed us to demonstrate its superiority compared to the first one.
We also worked on the intensity detection in opinion analysis. The study of scientific works and disagreements between corpora annotators showed us the subjectivity of this notion. Nevertheless, we have produced several models based on different corpora, which will serve as a basis for the project under development.
Finally, we participated in several Eloquant projects, such as the analysis of phone conversations, semantic enrichment for a Luxury House, the recognition of software names in short verbatims and the post-editing of verbatims translated from French into Russian. These projects helped us become familiar with the technologies used within the company.
Document type :
Master Thesis
Complete list of metadatas

https://dumas.ccsd.cnrs.fr/dumas-03081467
Contributor : Uga - Bulles <>
Submitted on : Friday, December 18, 2020 - 10:49:39 AM
Last modification on : Sunday, December 20, 2020 - 3:22:29 AM

Identifiers

  • HAL Id : dumas-03081467, version 1

Citation

Nadezhda Rumiantceva. Stratégies d'enrichissement sémantique d'analyse d'opinions : contributions logicielles et linguistiques. Sciences de l'Homme et Société. 2020. ⟨dumas-03081467⟩

Share

Metrics

Record views

47

Files downloads

21