USING ANIMATED MOOD PICTURES IN MUSIC RECOMMENDATION - Information Visualisation: Techniques, Usability and Evaluation - page 203

Metadata in MIR
Music Information Retrieval (MIR) is an interdisciplinary research field
studying different strategies for enabling access to new and historical
music collections. MIR includes diverse topics such as computational
methods for music analysis, human computer interaction (HCI), user
interface (UI) development, musicology, and music theory (Casey et al.
2008).
Typical MIR use cases include, for example, automatic classification
and transcription of music, making recommendations based on a given
seed song or personal profile information, providing metadata about an
unknown track, and finding music belonging to a certain genre or having a
certain mood. There are three strategies for solving different use cases: 1)
low-level audio features, 2) high-level music content description, and 3)
conceptual metadata. However, these strategies can also support each
other: musical tempo (a high-level music content descriptor) can be stored
as textual metadata inside the song. (Casey et al. 2008).
Currently, the most common way to access music collections is
through textual metadata describing the contents of the music collection
(Casey et al., 2008). Such metadata may include several keywords or
attributes (e.g., artist name, song name, mood, genre, tempo, or release
year of music) from a controlled vocabulary and their values (e.g., metal,
2012, or aggressive). The best-known metadata standard for music is ID3
(2012), which stores the metadata in the same files as the actual content.
Metadata for individual tracks in a music library can be determined
using content-based audio analysis (see e.g., [Dunker et al. 2008]), human
annotation, or hybrid approaches (e.g., Hu 2009). Human-annotated mood
information can be retrieved from various sources, including expert-
annotated labels (e.g., [AllMusic 2012]), social tags (e.g., [Lasf.fm 2012]),
annotation games (e.g., MoodSwings [Kim, Schmidt, & Emelle 2008]),
web page content, and lyrics.
Music discovery
One common MIR use case is the problem of music discovery, i.e., finding
music that is new to the listener. Music discovery can be active,
exploratory, or passive (Lillie 2008, p. 24).
In active discovery , the user has at least a rough idea of what he/she
wants; thus, he/she can search the music collection using artist names,
album names, or other type of metadata. In exploratory discovery , the user
browses through the music collection without knowing exactly what