State-of-the-Art: Semantics Acquisition Games - Semantic Acquisition Games: Harnessing Manpower for Creating Semantics - page 38

Fig. 3.3 Screenshot of the Peekaboom game, with purpose of identification of precise area that
object occupy in the image
to symbols (e.g. a mood information about a stream of individual notes). Therefore,
besides other crowdsourcing approaches, SAGs come to solve this task.
Some SAGs focus on sounds in general, other focus exclusively on music tracks,
as these represent an important resource segment in information retrieval. As we
mentioned in previous chapter, the spectrum of possible metadata types is wide, so
SAGs for describing music can also be differentiated according to this.
A first example of a music metadata acquisition game is the TagATune , created
by Law et al. [ 11 ]. In this game, two players hear a music track. The track may be
the same for both players, or they hear two different tracks. Their task is to describe
the track they hear and then decide, whether they are hearing to the same sample or
not. After the game, the descriptions are collected and processed for tags (or other
metadata).
An approach similar to the ESP game, but for music metadata, was created by
Mandel and Ellis [ 12 ]. These authors focused exclusively on musical resources,
presented to the players in 10 s slices. Their aim was to use different slices of the
same song to assess changes in the music of the same track. They also devised
a special scoring approach which encouraged players to enter specific tags about the
song they hear: the player scores only when he enters a tag which was entered by
another player exactly once (the game is single player). This forces the players not
to enter “the obvious” tags, but rather think about more specific and rare ones (but
of course, still relevant to the track). The ESP-like games on the other hand, use
a word banning mechanism, which a priori declares which tags are forbidden to use
for a particular image.