Considerations for Face-based Data Estimates: Affect Reactions to Videos

摘要

Video streaming is becoming the new standard for watching videos, providing an opportunity for affective video recommendation that leverages noninvasive sensing data from viewers to suggest content. Face-based data has the distinct advantage that it can be collected noninvasively with minimal equipment such as a simple webcam. Face recordings can be used for estimating individuals' emotional states based on their facial movements and also for estimating pulse as a signal for emotional reactions. We provide a focused case-based contribution by reporting on methodological challenges experienced in a research study with face-based data estimates which are then used in predicting affective reactions. We build on lessons learned to formulate a set of recommendations that can be useful for continued work towards affective video recommendation.