Little Search Game: Lightweight Domain Modeling - Semantic Acquisition Games: Harnessing Manpower for Creating Semantics - page 59

We were also able to fulfill the second prerequisite. Our software engineering
students were at the same the target player group—they represented the “initiated”
people, who have enough skill to perform well (and thus useful) in a word game in
a specialized domain.

4.4.1 TermBlaster Description

We now have the underlying document corpus changed. The rest of the changes made
in TermBlaster is oriented on the user interface of the game. The search queries,
resp. the negative search terms are no longer entered by player as a free text, but the
player picks them from a set of options. He does so by clicking on the bubbles with
words, spread around two-dimensional area (as seen on the screenshot in the Fig. 4.7 ).
We made this change to eliminate typing in the game, because it was identified by
some LSG players as an interface drawback of the game, which decrease the game's
dynamics.
The task for the player is formulated differently. Instead of LSG's “create a query
with lowest possible result count” the TermBlaster asks the player to simply “select
words that are mostly related to the given word (in the domain of software engineer-
ing)”. For each word, the player receives a reward in points, proportional to the real
relative co-occurrence of the given task word and selected word. This corresponds
to the original LSG principle: the more are the two terms co-occurring (relatively),
the more points the player receives.
Formally, let denote the domain term universum (set of

terms of the domain). Then, the player solves a task defined by task term q

Ω ={ ω 1 ,ω 2 , ..., ω n }

∈ Ω and

a set of “option terms” T ={ t 1 , t 2 , ..., t m }∈ Ω . The player consecutively selects

k terms from T to form A T . The score for each selected term

a i is computed from its relative co-occurrence with the task term. Formally, let R t
denote a result set yielded for a search query containing only term t . A partial score
s i for selecting i -th term is then defined as

={ a 1 , a 2 , .., a k }∈

= |

R a i ∩ R q |

s i

| R q |

where R a i denotes the result set for selected term and R q the result set for task term.
The total score s for all selected terms is then defined as

k

p |

R a i ∩ R q |

s = + c r

| R q |

i = 1

where p represents a denormalization constant ( 100) to make the point count larger

andwhere c r represents a smaller randomnumber (

≈

10) for score blurring and adding
some score for terms even not intersecting with the task term.

≈