Affective Metadata from Movies Helps Predict a Viewer’s Mood for Mood-based Recommenders
Finding a good movie to watch is getting more difficult in this day and age because of the huge availability. Movie recommendations are thus getting more important. Mood-based and context-aware recommenders could help improve these by taking a viewer’s mood and context into account. In this paper, we constructed affective ‘profiles’ for movies that potentially characterize their affective influence on viewer’s moods. We then gathered data about people’s moods before and after watching movies to use these affective profiles in predicting viewer’s moods after the movie. We found that a linear model armed with these affective profiles and information about people’s mood and an estimated score for the movie could explain 22% of the variance in moods. While not perfect, we believe this shows that affective movie profiles based on metadata combined with traditional recommender systems could successfully use people’s moods to improve movie recommendations.
Tijl Kindt, "Affective Metadata from Movies Helps Predict a Viewer’s Mood for Mood-based Recommenders", Master's Thesis for the Media Technology programme, Leiden University (The Netherlands), 2013