Affivir: An affect-based Internet video recommendation system

In this paper, we present Affivir, a video browing system that recommends Internet videos that match a user's affective preference. Affivir models a user's watching behavior as sessions, and dynamically adjusts session parameters to cater to the user's current mood. In each session, Affivir discovers a user's affective preference through user interactions, such as watching or skipping videos. Affivir uses video affective features (motion, shot change rate, sound energy, and audio pitch average) to retrieve videos that have similar affective responses. To efficiently search videos of interest from our video repository, all videos in the repository are pre-processed and clustered. Our experimental results show that Affivir has made a significant improvement in user satisfaction and enjoyment, compared with several other popular baseline approaches.