LA/Opt Seminar: How do we feel about what we see? Multimodal Machine Learning based models for recognizing induced emotions of movie audiences, Michal Muszyinski, University of Geneva

Linear Algebra and Optimization Seminar (CME 510)
ICME, Stanford University

Title: How do we feel about what we see? Multimodal Machine Learning based models for recognizing induced emotions of movie audiences
Speaker: Michal Muszyinski, University of Geneva
Date: November 2, 2017
Time: 4:30pm
Location: Y2E2 111

Abstract:

Affective computing as a modern branch of computer science develops
systems and devices that can process, recognise and interpret human
affects. Predicting emotional responses of movie audiences to
affective movie content is a challenging task in affective computing.
Previous work has only focused on using audio-visual movie content to
predict induced emotions of movie audiences. However, a relationship
between audiences' perceptions of affective movie content (perceived
emotions) and emotions evoked in audiences (induced emotions) remains
unexplored.

In this work, we address the relationship between perceived and
induced emotions of movie audiences, and identify features and
modelling approaches effective for predicting induced emotions. We
extend the LIRIS-ACCEDE database by annotating perceived emotions in a
crowd-sourced manner, and discover that perceived and induced emotions
are not always consistent. Furthermore, we find that affective
cue-based features and movie audiences' physiological and behavioral
reactions are effective predictors of induced emotions. Our
experiments show that induced emotion recognition can benefit from
different architectures of machine learning models that include
temporal context and multimodal information. We test three fusion
strategies of multimodal information: feature level, decision level
and hierarchical fusion. Our study bridges the gap between affective
content analysis and induced emotion prediction.

Date: 
Thursday, November 2, 2017 - 4:30pm to 5:30pm
location: 
Jerry Yang and Akiko Yamazaki Environment and Energy Building (Y2E2), 473 Via Ortega, Stanford, CA 94305, USA