DynAMoS Database

Dynamic Affective Movie Clip
Database for Subjectivity Analysis

ACII 2023 | Girard, Tie, & Liebenthal
https://dynamos.mgb.org

Types of Emotion Ratings

  • Who is providing the measure of emotion?
    • Self-reports of participants’ own emotion
    • Observer-perceptions of others’ emotion
  • How is emotion being represented?
    • Discrete choices of emotion categories (e.g, happy, angry)
    • Continuous scores on emotion dimensions (e.g., valence)
  • How often are the emotions reported?
    • Holistic ratings collected once after each stimulus
    • Dynamic ratings collected repeatedly during each stimulus

Ambiguity and Subjectivity

  • Emotion ratings will inevitably vary between raters…

  • We usually treat such variability as a nuisance to “fix”

  • But we can gain a lot by embracing these differences: studying their sources and building them into our models

  • Ambiguity: variability across different observers’ perceptions of the emotion in a given stimulus (see Sethu et al., 2019)

  • Subjectivity: variability across different subjects’ self-reports of the emotion they experienced from a given stimulus

DynAMoS Database

  • Participants
    • Healthy community members from Rally with MGB
    • 83 participants (67% female, 18–59 years old)
    • 43 White, 22 Asian, 12 Black, 5 Other; 11 Hispanic/Latino
  • Procedure
    • Watch 22 affective movie clips (2.2–7.1 minutes each)
    • Dynamic self-reported emotional valence (CARMA)
    • Holistic self-reported positive/negative affect (S-PANAS)

Emotional Ratings

Dynamic

–4 (negative) to +4 (positive)
Rated at 30 Hz, binned to 1 Hz

Holistic

Each rated 0 (very slightly or not at all) to 4 (extremely):

Positive Affect
alert, determined, enthusiastic, excited, inspired

Negative Affect
afraid, distressed, nervous, upset, scared

https://carma.jmgirard.com

Quantifying Subjectivity

Estimates from a Bayesian G study (github.com/jmgirard/varde)

Visualizing many time series is hard…

Each blue line is one participant’s time series

Introducing the Chromodoris plot

Black = Mean, Yellow = 50%, Green = 70%, Purple = 90%

Database Uses

  • Emotion elicitation video set with normative data

  • Affective content analysis with average ratings

  • Subjectivity analysis to predict rating distributions

  • Subjectivity analysis to explain degree of subjectivity

  • Personalized modeling of affective reactions

Future Directions

  1. Add more movie clips and participants

  2. Add more dynamic and holistic rating dimensions

  3. Add data from sensors (e.g., physiological and eye tracking)

  4. Collect information about participants’ personality

  5. Collect similar data in clinical/medical populations

Acknowledgements

  • Co-authors
    • Jeffrey Girard (University of Kansas)
    • Yanmei Tie (Brigham & Women’s Hospital, HMS)
    • Einat Liebenthal (McLean Hospital, HMS)
  • Funding
    • Alexandra Golby (Brigham & Women’s Hospital, HMS)
  • Assistance
    • Colin Gavin, Laura Rigolo, Abby Recko, Ben Phan

Questions?

https://dynamos.mgb.org