State-of-the-Art: Semantics Acquisition Games - Semantic Acquisition Games: Harnessing Manpower for Creating Semantics - page 46

are dealing with. Mostly, these involve semantics acquisition: mainly multimedia
description and domain modeling. Most of the existing approaches deliver, with
high probability, the semantics with acceptable quality.
We immersed into the existing SAGs solutions from their purpose perspective
(what semantics they acquire and how). For now, we left aside the design aspects
of these games (although the reader might already noticed certain recurring patterns
and features of the SAGs)—we cover them in the second part of this topic. From
the purpose point of view, the existing SAGs involve mainly creation of lightweight
semantics: either description of resources with simple tags or assessing triplets under
simple schemes. Only few SAGs participate in the “heavy” semantics acquisition.
Still, the lightweight SAG products are useful in today's Web (which welcomes any
semantics available), and so, the SAGs themselves have their place.
The existing SAG-based approaches have a certain limitation: they operate in
general domain (of images, music, texts, models). It is somewhat natural: the general
domain knowledge is known to many people. They can solve tasks related to it
(e.g. annotate a general image, with general metadata). This makes the potential
pool of players large, which means the games have larger quantitative potential
and can freely use the redundant player work. On the other hand, the general domain
represents only a fraction of areas where semantics are needed. These are the specific
domains which can comprise any problem areas which somehow need to be modeled:
frompersonal images, through very specificmusic genres tomedical knowledge. This
space is heterogeneous, but for each field, the following is valid: much less people
understand it and is able to perform tasks of creating semantics for it. Therefore, the
use of existing SAGs is seriously hindered here. We see major research opportunities
to develop SAG solutions that would overcome this barrier. They would probably
need a very specific designs, but we might also discover more widely usable design
patterns which can be used to push SAGs to specific problem areas. In this work, we
introduce two SAGs aimed towards acquisition of specific semantics: for creation of
specific domain model and for personal image metadata.
Another question is about the impact of the existing SAG solutions. Although
the existing solutions appear to cover pretty much of the semantics acquisition prob-
lems, some areas are still left untouched (such as problems with poor validity of
the existing metadata on the Web). Also, from the quantitative perspective, only
a few solutions [ 3 , 24 ] managed to quantitatively justify themselves in practice.
Other solutions have only a limited impact or remain in purely experimental condi-
tions. This could be possibly accounted to many reasons (e.g. cold start problems,
attractiveness, understandability), but these we analyze later. Nevertheless, the exist-
ing issues justify the efforts for creating new SAG solutions, which can possibly be
more effective.