Supervision : Maria RIFQI

Co-supervision : LESOT Marie-Jeanne, MARSALA Christophe

Representation and learning for both emotional and dynamic information from texts.

Automatic knowledge extraction from texts consists in mapping low level information, as carried by the words and phrases extracted from documents, to higher level information. The choice of data representation for describing documents is, thus, essential and the definition of a learning algorithm is subject to their specifics. This thesis addresses these two issues in the context of emotional information on the one hand and dynamic information on the other.
In the first part, we consider the task of emotion extraction for which the semantic gap is wider than it is with more traditional thematic information. Therefore, we propose to study representations aimed at modeling the many nuances of natural language used for describing emotional, hence subjective, information. Furthermore, we propose to study the integration of semantic knowledge which provides, from a characterization perspective, support for extracting the emotional content of documents and, from a prediction perspective, assistance to the learning algorithm.
In the second part, we study information dynamics: any corpus of documents published over the Internet can be associated to sources in perpetual activity which exchange information in a continuous movement. We explore three main lines of work: automatically identified sources; the communities they form in a dynamic and very sparse description space; and the noteworthy themes they develop. For each we propose original extraction methods which we apply to a corpus of real data we have collected from information streams over the Internet.

2010-2014 Publications

• 2014
  • F. Dzogang, M.‑J. Lesot, M. 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, vol. 28 (1), Affects, compagnons artificiels et interactions, pp. 131-157, (Editions Hermes) (2014)
• 2013
  • F. Dzogang : “Représentation et apprentissage à partir de textes pour des informations émotionnelles et pour des informations dynamiques”, thesis, phd defence 07/18/2013, supervision Rifqi, Maria, co-supervision : Lesot, Marie-Jeanne, Marsala, Christophe (2013)
• 2012
  • F. Dzogang, M.‑J. Lesot, M. Rifqi : “Fusion anticipée de descripteurs bas niveau pour la détection d’émotions dans les textes”, Workshop Affect, Compagnon Artificiel, Interaction, WACAI'12, Grenoble, France (2012)
  • F. Dzogang, M.‑J. Lesot, M. Rifqi, B. Bouchon‑Meunier : “Early Fusion of Low Level Features for emotion Mining”, Biomedical Informatics Insights, vol. 5, pp. 129-136 (2012)
  • F. Dzogang, Ch. Marsala, M.‑J. Lesot, M. Rifqi : “An Ellipsoidal K-Means for Document Clustering”, IEEE 12^th International Conference on Data Mining (ICDM 2012), Bruxelles, Belgium, pp. 221-230, (IEEE) (2012)
• 2010
  • F. Dzogang, M.‑J. Lesot, M. Rifqi, B. Bouchon‑Meunier : “Expressions of Graduality for Sentiments Analysis - A Survey”, IEEE International Conference on Fuzzy Systems, Fuzz'10, Barcelona, Spain, pp. 1394-1400, (IEEE) (2010)
  • F. Dzogang, M.‑J. Lesot, M. Rifqi, B. Bouchon‑Meunier : “Analysis of texts’ emotional content in a multidimensional space”, International Conference on Kansei Engineering and Emotional Research KEER 2010, Paris, France (2010)