Automated prediction of preferences using facial expressions

Area of Science:

• Computer Vision
• Social Cognition
• Affective Computing

Background:

• Automated analysis of facial expressions is crucial for understanding social cognition.
• Predicting human preferences from non-verbal cues remains a significant challenge.

Purpose of the Study:

• To introduce and address the problem of automated attitude detection from spontaneous facial expressions using computer vision.
• To develop and evaluate algorithms for predicting human preferences based on facial dynamics.

Main Methods:

• Facial landmark detection from covert video recordings.
• Temporal dynamics analysis of facial landmarks to predict preferences.
• Development of two algorithms for preference prediction.

Main Results:

• The developed algorithms achieved accuracy comparable to human judges in predicting preferences.
• Despite high accuracy, algorithms were not perfect, indicating room for improvement.
• The study utilized a database of 768 videos across four preference types.

Conclusions:

• Automated attitude detection from facial expressions is feasible with computer vision techniques.
• Facial dynamics offer valuable information for predicting human preferences.
• The public release of the dataset will facilitate further research in this domain.