A real-world dataset of group emotion experiences based on physiological data

Area of Science:

• Affective computing
• Physiological signal processing
• Human-computer interaction

Background:

• Affective computing excels at facial emotion recognition but struggles with integrating physiological data.
• Existing emotion recognition models using physiological data perform well in labs but not in real-world scenarios.
• There is a need for large-scale, real-world datasets to improve the generalizability of physiological emotion recognition.

Purpose of the Study:

• To introduce G-REx, a novel dataset designed for real-world affective computing.
• To enable the development of robust emotion recognition systems using physiological data collected in naturalistic settings.
• To facilitate research on emotion recognition in group contexts.

Main Methods:

• Collected photoplethysmography (PPG) and electrodermal activity (EDA) using wrist-worn devices.
• Acquired data during long-duration movie sessions in a group setting.
• Performed retrospective emotion annotation on segments with significant physiological responses.

Main Results:

• The G-REx dataset comprises over 380 hours of data from 190+ subjects across 31+ movie sessions.
• Data collection focused on real-world conditions, mimicking everyday scenarios.
• The dataset's group setting provides contextual information for emotion recognition.

Conclusions:

• G-REx offers a valuable resource for advancing real-world affective computing.
• The dataset's design promotes easy replication, encouraging further large-scale data collection.
• This work addresses the gap in generalizing physiological emotion recognition to everyday life.