Automatic Evaluation of Video Contents Using Eye Gaze and Facial Expression
Ryuzo Ishikawa, Hiroki Nomiya · 2024
The rating value of video contents is important for creators and viewers. It is difficult to obtain accurate evaluations using current evaluation methods such as good buttons and 5-level evaluations. The main reason is that there are many viewers who do not rate videos due to tedious task. To make the evaluation of videos more reliable and accurate, we propose a method to create features from the eye gaze and facial expressions of video viewers and automatically appreciate video contents using a machine learning model. The accuracy on a 5-level evaluation is 46.8% for all participants' data and 52.7% for those who expressed facial expressions while watching the video. Although it is difficult to perform evaluations using only eye gaze, it is suggested that the accuracy could be improved by adding eye gaze to facial expressions.