CityLights: A Method for Music Metadata Validation - Semantic Acquisition Games: Harnessing Manpower for Creating Semantics - page 78

Chapter 6
CityLights: A Method for Music Metadata
Validation
Abstract This chapter presents our semantics acquisition game called CityLights
and experiments related to it. It has the purpose of validating of existing tags for
musical resources in datasets where tag quality (correctness, descriptiveness) is not
ensured. The game has a form of music quiz, where player first hears a music track
and then he is asked to select one of the prepared tag sets, which best characterizes
the music track. The tags that mostly confuse players (lead them to false answers)
are declared invalid (and are removed from the dataset).
A general need for metadata exists also in the domain of music multimedia
resources. Music metadata comprises a wide range of types such as author and
interpret of the music, creation or recording year, genre, mood, lyrics theme, place
where the music should be listened, rhythm, tonality etc. Some of these metadata
are generally provided by the resource creators, other may be inferred automatically
[ 1 , 4 ], but as with many other multimedia, a human (crowd) labor [ 2 ] is needed to
acquire all types of metadata for music tracks.
The existing approaches result into a solid quantity of metadata available for musi-
cal resources. However, the quality of these metadata is unreliable. Users introduce
a lot of noise to the tags, contributing with subjective descriptions and sentiment.
They may also assign wrong tags (such as genre). A simple crowdsourcing princi-
ple: a cross-user validation results in a very limited set of metadata (mostly tags).
A noise is also introduced by automated metadata extraction methods. During our
experiments with the LastFM corpus, we have identified that more than a half of its
tags was assigned wrongly or with very weak connection to the music tracks.
In such conditions, the music metadata acquisition problem turns into the vali-
dation problem: rather than creating new metadata, we need to validate the existing
sets to re-rank them and filter out those incorrectly assigned.
The music metadata validation is the task for our method and a SAG called
CityLights [ 3 ]. The game works with real data corpus—the LastFM tag database.
The authors wish to thank the bachelor student, Peter Dulacka, for participation on the design,
implementation and validation of the game