Game Development Reference
In-Depth Information
Fig. 4.2 LSG-created term
network subset
of results. The player may also observe this on the chart, showing history of the
attempts.
3. The player may add more negative search terms. He may also start to change
the search terms for other, to refine the query and better his results. The player
may do as many attempts as he wants (until he is satisfied with his score gain,
which he may review in the ladder on the right). There is, however, a limit of
N negative terms (in the deployed game version, we used 6), which can be used
at the same time. The number must naturally be limited, because with infinite
number of terms, the game would be trivial. On the other hand, we wanted to
provide some “maneuvering” space for the player, so he also could experiment
with particular term combinations.
4. The game ends, when player submits the current score. He may then review
overall ladder (combining results from all games played) or play another game
round (with different or same term).
The winning condition—the lowest possible result count yielded—forces players to
enter negative terms that have high co-occurrence with the task term on the Web,
which is, due to the tendency of humans to think about the concepts, interpreted as
relatedness of those terms. After several games were played on the same task, the
agreement principle to validate answers can be applied.
4.2 Term Relationship Inference
The game rules (sourcing from the principles of negative search) force players to
disclose their opinions on the relatedness of terms. When some term-term combina-
tions are suggested by multiple players, they may become a part of term relationship
network such as one depicted in the Fig. 4.2 . In this section we describe the network
construction, out of the raw logs of the game.
The subgroup of game logs relevant for network construction can be represented
as “triplets”
(
p i ,
t j ,
N ij )
where
p i ;
p i
P denotes player identifier from the set of all players P .
t j ;
t j
T denotes task term from the set of all task terms T .
N ij ={
denotes a set of negative terms used at least once by player
p i for task t j .Alsolet
n ij 1 , ...,
n ijn }
n
;
n
N where N is the set of all negative words.
These “triplets” constitute a basis for termnetwork filtering. From them, we construct
the set of term combination suggestions of a certain player, called votes (each vote is
denoted as l
= (
p i ,
t j ,
n k )
:
×
×
of vote set L
P
T
N ).We do not consider information
 
Search WWH ::




Custom Search