Analyse automatique des données sociales (Social Computing)

Ce thème a pour vocation de rassembler les recherches autour des modèles computationnels pour l’analyse des interactions sociales que ce soit pour l’analyse du web ou la robotique sociale. La particularité de ce thème est sa pluridisciplinarité : les modèles computationnels sont établis en étroite collaboration avec les domaines de recherche tels que la psychologie, la sociologie, et la linguistique. Ils reposent sur des méthodes issues de domaines variés en traitement du signal (ex: traitement du signal de la parole pour la reconnaissance d’émotions), en apprentissage automatique (ex : utilisation des Champs Aléatoires Conditionnels pour la détection d’opinions dans les textes), en informatique ( ex : le traitement automatique du langage naturel pour la détection d’opinions, la prise en compte de la composante socio-émotionnelle dans les interactions humain-machine).

Les recherches menées sur ce thème au sein du département TSI s’articulent autour des trois items suivants :

  • l’analyse des données sociales du web
    • les données textuelles : l’opinion mining, argument mining
    • les données relationnelles : l’analyse des graphes de communauté, l’analyse des données de préférences, l’analyse des logs
  • le traitements des signaux sociaux (social signal processing) et la reconnaissance d’émotions (analyse multimodale – texte, audio, vidéo, des signaux sociaux et émotionnels) dans les interactions humain-humain et humain-agent.
    • Détection des opinions et des comportements socio-émotionnels de l’utilisateur
    • Reconnaissance des émotions dans l’écriture manuscrite
    • Génération de comportements socio-émotionnels
  • les stratégies socio-émotionnelles d’interaction et dialogue humain-machine

Les applications de ce thème incluent : l’e-reputation, les systèmes de recommandation,  la fouille de données, la robotique sociale,  les agents conversationnels animés et les compagnons virtuels (Cortana, Google now, Siri), l’indexation multimedia, les applications médicales (aide au diagnostique).

Thème transdisciplinaire, en lien les activités du département Sciences Economiques et Sociales :

Enseignants-chercheurs département Image Données Signal : Chloé Clavel, Florence d’Alché Buc, Slim Essid, Laurence Likforman, Giovanna Varni, Anne Sabourin

Enseignants-chercheurs département Informatique et réseaux : Jean-Louis Dessalles

Post-doctorants : Atef Ben Youssef, Varun Jain, Guillaume Dubuisson

Doctorants : Caroline Langlet, Thomas Janssoone, Irina Poltavchenko, Valentin Barrière, Alexandre Garcia

Contact : Chloé Clavel

Prochain séminaire : 

Date: Friday 6 July at 9:30

Place : Amphithéâtre Jade, telecom-paristech, 46 Rue Barrault, 75013 Paris, Metro Corvisart/Placed’Italie

Title: Identifying High Quality Arguments and Argument Facets in Social Media

Online argumentative dialogue is a rich source of information about
popular beliefs and opinions that could be useful to companies as well
as governmental or public policy agencies. Compact, easy to read,
summaries of these dialogues would thus be highly valuable, but a
priori it is not even clear what form such a summary should take. This
lecture will describe methods for processing online dialogues in order
to automatically identify high quality arguments, along with induction
of and identification of each topic’s argument facets. We explain
these techniques can be used to produce dialogic summaries, and also
show how we have used them to develop a chatbot that can argue about
current social and political topics.
Bio : Marilyn Walker, is a Professor of Computer Science at UC Santa Cruz,
and a fellow of the Association for Computational Linguistics (ACL),
in recognition of her for fundamental contributions to statistical
methods for dialog optimization, to centering theory, and to
expressive generation for dialog. Her H-index a measure of research
impact is 55. Her current research includes work on computational
models of dialogue interaction and conversational agents, analysis of
affect, sarcasm and other social phenomena in social media dialogue,
acquiring causal knowledge from text, conversational summarization,
interactive story and narrative generation, and statistical methods
for training the dialogue manager and the language generation engine
for dialogue systems. Before coming to Santa Cruz in 2009, Walker was
a professor of computer science at the University of Sheffield. From
1996 to 2003, she was a principal member of the research staff at AT&T
Bell Labs and AT&T Research, where she worked on the AT&T Communicator
project, developing a new architecture for spoken dialogue systems and
statistical methods for dialogue management and generation. Walker has
published more than 200 papers and has 10 U.S. patents granted. She
earned a B.A. in Computer and Information science at UC Santa Cruz,
M.S. in Computer science at Stanford University, and a M.A. in
Linguistics and Ph.D. in Computer Science at the University of


Séminaires passés :

Date : mercredi 2 mai à 12h
Lieu : salle B547

Titre : State-dependent valuation and confirmatory biases in human reinforcement learning

Intervenant : Stefan Palminteri, Human Reinforcement Learning team, Institut National de la Recherche Médical & Ecole Normale Supérieure

Résumé : Valence is a fundamental concept in the learning, decision-making and emotion literature. However, in the reinforcement learning context, different facets of this psychological concept are often confounded and/or ill-defined. In the present talk we theoretically propose, mathematically formalize and experimentally investigate two different facets of valence: ‘contextual’ and ‘informational’. Combining imaging and modeling techniques in a series of studies, we found that these different aspects have dissociable computational and neural correlates, and exert powerful effects on human learning behavior.
Références principales :

Palminteri S, Justo D, Jauffret C, Pavlicek B, Dauta A, Delmaire C, Czernecki V, Karachi K, Capelle L, Durr A, Pessiglione M. Critical roles for anterior insula and dorsal striatum in punishment-based avoidance learning. Neuron (2012).

Palminteri S, Khamassi M, Joffily M, Coricelli G. Contextual modulation of value signals in reward and punishment learning. Nature Communications (2015).

Lefebvre G, Lebreton M, Meyniel F, Bourgeois-Gironde S, Palminteri S. Behavioural and neural characterization of optimistic reinforcement learning. Nature Human Behaviour (2017).


Date : Mercredi 21 mars à 12h en B547

Title : Pauses in the perception and production of fluent speech: The case of nonnative listeners and speakers

Abstract: The production of silent and filled pauses in speech is known to vary by language. If production proceeds and motivates perception, then it is expected that listeners—including nonnative listeners—will show variation in how pauses influence their perception of speech. In this talk, I will share results from an experimental evaluation of this prediction with several different language groups.  Although still in progress, results support the hypothesis and show that nonnative listeners are sensitive to both silent and filled pauses, but in slightly different degrees.

Based on these results, I will discuss how nonnative speakers may become more aware of pauses in their own speech and how it relates to their second language fluency.  I will introduce and demonstrate Fluidity—an application designed to help learners increase their speed fluency. The application gives real-time feedback on such fluency-related measures as pauses via an expressive virtual agent. Some results from usability testing will be presented in order to show how the application is received by learners and their suggestions for future development.

Bio: Ralph Rose is a member of the Faculty of Science and Engineering at Waseda University in Tokyo, Japan where he and his colleagues manage the English language program. His research interests include filled pauses (uh/um), second language fluency, and educational technology—and particularly the intersection of these three.