Game Development Reference
In-Depth Information
However, the original tag extraction procedure rely heavily on the cross-player
validation. It needs a substantial number of players playing the game with the same
data set. But such pool of players is not available in this case, because the specificness
of the desired metadata narrows the group of people able to provide them. This is
an disadvantage of any crowdsourcing process aiming for a specific information
acquisition. Overcoming it may include increasing worker motivation and finding
alternative ways to validate the produced information. Our solution utilizes both.
Concerningmotivation to play, the PexAce-Personal comprises threemotivational
aspects:
1. The original game challenge (e.g. finish the game, beat the highest score). Inter-
estingly, even when the players play with different image sets, their mutual com-
parison according to score is possible, because the scoring is in fact independent
on the actual image content.
2. Interaction with familiar content. While playing, the players experience the plea-
sure of reviewing their images, especially those they have not seen for longer
time.
3. The players that are aware of the game's purpose know that they are helping
themselves or their relatives or friends. This strengthens their motivation for
participation.
Motivated players, aware of the game's purpose would produce more accurate and
specific annotations. This allows us to rely more on the sole player tag suggestions
and lowers the need for cross-player tag validation. This is very convenient for the
use case, because we can expect that in some cases, only one player will be available
to play with one image set (although the game encourages player to “share” it among
friends, to “help him by playing”). Thus, the images used in the game are personal
images of the player or belong to his social circle.
The original cross-player tag validation was complemented with several less
restrictive heuristics enabling the game to accept tags entered by only one player.
They also optionally exploit a common album-like groupings of images in personal
repositories.
The pre-processing of the annotations created in the PexAce-Personal is executed
in the samemanner aswith the original game. The texts are tokenized and lemmatized,
the stopwords are removed. The only difference is the missing automated translation
to English, which we omitted in this case, since it would be too inaccurate for specific
texts (often containing metaphorical references). The natural language processing
procedures were therefore implemented for each language separately and the players
were asked to explicitly select their game language (for the experiments described
below, we used Slovak language).
As with the original approach, the output of pre-processing were tag sugges-
tions triples—
player-image-tag
. In PexAce-Personal, we add a fourth element to the
suggestion, the identifier of
album
, in which the image was present (an informa-
tion often available). When the album information is not available, the validation
heuristics utilizing it are not applied. The set of the tag suggestion quartets (player-
image-tag-album) are subject to the following filtering heuristics:
Search WWH ::
Custom Search