State-of-the-Art: Semantics Acquisition and Crowdsourcing - Semantic Acquisition Games: Harnessing Manpower for Creating Semantics - page 20

other hand, experts are sometimes unavailable. Second, lay contributors are much
cheaper or even free (resp. they are not paid for solving a task).

2.5.1 Crowdsourcing Classifications

In general, crowdsourcing comes in many different flavors. It also has very strong
overlap with other terms such a human computation, social computing, collective
intelligence, crowd computing. Together, they comprise a loosely bounded field and
several researchers reflected the lack of abstract, formal models describing it. This
resulted into several survey publications, attempting to conceptualize the field with
variety of classifications [ 19 , 21 , 23 , 54 , 55 ].
In his position paper, Erickson classifies the crowdsourcing systems according to
distribution of the crowd in time and space [ 23 ] (being either at the same time/place
or not). This results into four categories of crowdsourcing:
￿ Audience based , when entire crowd participates at the same time and space.

￿ Event based , when the crowd is geographically distributed, but works at the same

time on a common goal (e.g., innovation competition).

￿ Geocentric , when the work is done at a particular geographical location bymultiple

workers in different times (e.g., communal problem reporting).

￿ Global , when the process is bound neither to time nor space.

The typical Crowd-based semantics acquisition approaches (e.g., semantics acqui-
sition games) are found in the latter category (global), as there is usually no need to
bound them to specific time or place (though as a significant exception, various geo-
centric applications for collecting metadata on points of interest, e.g., FourSquare,
should be mentioned).
Doan et al. defines nine dimensions according to which the crowdsourcing appli-
cations could be considered [ 21 ]. We look at six of these dimensions interesting
through prism of crowd-based semantics acquisition:
￿ What type of target problem is being solved (e.g., labeling images, building

a knowledge base, rating movies)?

￿ What is the nature of collaboration? Authors identify two major groups of

approaches: explicit (where users explicitly collaborate to create an useful arti-
fact as their primary objective) and implicit (where users solve a target problem as
a side effect of another activity). The semantics acquisition tasks fall in both cate-
gories. The explicit approaches include item rating or knowledge base (ontology)
building, the implicit comprise for example, image tagging through crowdsourcing
games.

￿ How does the application recruit and retain new workers? This perspective brings

up the question of incentives to the workers. Some semantics acquisition applica-
tions are useful for the user himself (e.g., tagging websites in bookmark portal),
some rely on volunteers (e.g., contribution to knowledge bases like ConceptNet),
some motivate by entertainment (e.g., crowdsourcing games).