USING ANIMATED MOOD PICTURES IN MUSIC RECOMMENDATION - Information Visualisation: Techniques, Usability and Evaluation - page 204

he/she is looking for. Still, the user can get satisfaction from the browsing
experience itself.
Passive discovery is related to the use of music recommendation
systems, which push content and recommendations to the user which are
based on a personal preference profile or one or more example (seed)
songs. In the first case, the system observes the personal preferences of
users, models the properties of a music catalogue, and then suggests music
based on the model. In the latter case, the user either selects the seed song
by him/herself, or adjusts some parameters (e.g., mood or genre) on the
basis of which the seed songs are selected, and the recommendations are
then generated. For more details on the different types of music
recommendation systems, see, for example, Celma 2008.
Categorizing music using moods
Listening to music can affect a person's mood, facial expressions and
physiological reactions. As a consequence, music is often used for mood
enhancement and relaxation, motivation during sports exercises, and
moderating or boosting arousal levels. For example, “people in a state of
unpleasantly high arousal (e.g., while driving in heavy traffic) generally
prefer quiet, relaxing music, while people who are in a state of pleasantly
high arousal (e.g., while working out) will prefer loud, energizing music”
(Levitin 2007). As music can express, communicate, and evoke emotions
in listeners and performers (Juslin & Sloboda 2001), it is also natural for
humans to categorize music in terms of emotional reactions, emotions,
and/or moods.
Emotions differ from moods in three ways: 1) they last for a shorter
time, 2) they have an identifiable stimulus event (whereas moods do not),
and 3) they are accompanied by distinct facial expressions (whereas
moods are not) (Sloboda and Juslin 2001). However, the terms are often
intermixed in the MIR literature; both terms are also used in this chapter.
To classify emotions, MIR systems typically use categorical or
dimensional approaches, both with roots in psychological research.
Regardless of the approach used, the collection of “ground truth” emotion
labels remains a challenging problem (Kim et al. 2010).
The categorical approach is based on the concept of basic emotions,
meaning “there is a limited number of innate and universal emotion
categories from which all other emotional states can be derived” (Sloboda
& Juslin 2001, p. 76). Ekman (2004) proposed six emotions (anger,
disgust, fear, joy or happiness, sadness, and surprise) that are recognized
universally.