Apprentissage de concepts émotionnels à partir de descripteurs bas niveau

Abstract : This paper addresses the task of emotion recognition in unstructured textual documents. It first reviews existing representations of documents able to cope with the subjectivity of their emotional content. We then describe the proposed method: following an early fusion strategy, features defined as n-grams of several orders are combined. Moreover, dictionaries specific to each emotion label are automatically extracted. The proposed decision process is implemented as a two level "one vs. all" strategy relying on linear SVM. The resulting system has been applied to the I2B2 track2 challenge and obtained a good ranking among systems relying on a low level representation of the data. We detail the results obtained over the corpus made of real data describing 4 241 sentences labeled with 12 emotion labels and 3 non emotional labels.
Document type :
Journal articles
Complete list of metadatas

https://hal.archives-ouvertes.fr/hal-01078641
Contributor : Maria Rifqi <>
Submitted on : Tuesday, November 4, 2014 - 5:50:24 PM
Last modification on : Friday, May 24, 2019 - 5:29:17 PM

File

DzogangLesotRifqi_ria.pdf
Publisher files allowed on an open archive

Identifiers

Citation

Fabon Dzogang, Marie-Jeanne Lesot, Maria Rifqi. Apprentissage de concepts émotionnels à partir de descripteurs bas niveau. Revue des Sciences et Technologies de l'Information - Série RIA : Revue d'Intelligence Artificielle, Lavoisier, 2014, Affects, compagnons artificiels et interactions, 28 (1), pp.131-157. ⟨10.3166/ria.28.131-157⟩. ⟨hal-01078641⟩

Share

Metrics

Record views

258

Files downloads

250