Game Development Reference
In-Depth Information
decision when to suspend an image from occurring in the games. In our experiments,
we used a rather naive approach: an image is featured within the games until the
number of its annotations reaches 15 or number of its output tags reaches five.
Although this was sufficient for executing our experiments and for proving the
method concept, in practice, such approach introduces several types of events which
may occur during the annotation process and which cause that human work is used
inefficiently. They would mostly be caused by semantic characteristics of each indi-
vidual image, namely the number and relevancy of concepts related to it.
When an image has only few dominant concepts related to it, the players would
probably use only these in their annotations. This “low tag diversity” would cause
that many annotations would be unnecessary redundant (repeating the same tags)
and neither output tag count or annotation count thresholdwould prevent “wasting”
of the player “work” which could be used elsewhere.
When the number of concept related to the image is high (with no “apparently
dominant” concepts) the tag suggestions of the player crowd become more diverse.
This can cause a situation, whenmany “good” tag suggestions remain unconfirmed
by other players upon reaching the annotation count threshold.
In the future, a possible solution to this problems may lie in a more dynamic
approach based on some diversity measurement metric used for making decisions
about annotation suspension. Moreover, in case of “high tag diversity problem”
a new game mechanics could possibly be used: assuming that players suggest many
tags to an image (not yet validated by other players), the game could use these to
“auto-tag” the image for the player. This way, they could validate the existing tags
instead of creating new ones. The positive or negative outcome of the validation
would be determined by the player behavior: either he would make mistakes using
the auto-tagged image (which would indicate that the tag is wrong) or he would
successfully use it to find pairs (in this case, the featured tags would probably be
right).
References
1. Ke, X., Li, S., Cao, D.: A two-level model for automatic image annotation. Multimedia Tools
Appl.
61
, 195-212 (2011)
2. Šimko, J., Tvarožek, M., Bieliková, M.: Human computation: image metadata acquisition based
on a single-player annotation game. Int. J. Hum. Comput. Stud.
71
(10), 933-945 (2013)
3. Vainio, T., Väänänen-Vainio-Mattila, K., Kaakinen, A., Kärkkäinen, T., Lehikoinen, J.: User
needs for metadata management in mobile multimedia content services. In: Proceedings of the
6th International Conference on Mobile Technology, Application & Systems, Mobility '09,
pp. 51:1-51:8. ACM, New York (2009)
Search WWH ::
Custom Search