Exploiting social graph networks for emotion prediction.

Area of Science:

• Computational Social Science
• Affective Computing
• Mobile Health

Background:

• Emotion prediction is vital for mental healthcare and emotion-aware computing.
• Predicting emotions is challenging due to physiological, mental, and environmental influences.
• Existing methods often overlook the impact of social context and environmental factors.

Purpose of the Study:

• To predict self-reported happiness and stress levels using mobile sensing data.
• To incorporate environmental data (weather) and social network information into emotion prediction models.
• To develop a scalable machine learning architecture for dynamic social network integration in affect prediction.

Main Methods:

• Utilized mobile sensing data for emotion prediction.
• Constructed social networks from phone data without additional user cost or privacy concerns.
• Developed a graph neural network architecture to aggregate multi-user data and integrate temporal dynamics.
• Incorporated weather data alongside physiological and social network information.

Main Results:

• Demonstrated significant prediction performance improvement by integrating social network data.
• The proposed architecture effectively handles dynamic social network distributions and scales to large networks.
• Investigated the impact of graph topology on emotion prediction accuracy.

Conclusions:

• Integrating social network information derived from mobile sensing data enhances emotion prediction.
• The developed scalable architecture automates social network integration for affect prediction.
• This approach offers a privacy-preserving and cost-effective method for improving mental healthcare and emotion-aware computing.