Activity - Game Invaders: The Theory and Understanding of Computer Games - page 69

activities a user performs was not feasible. Recording activities automatically is still
a long way past technology's capabilities and employing a manual “ game tester ”
would be hugely expensive as well as time consuming.
Analyzing professional game reviews involved reading each review for every
game and analyzing the way in which professional reviewers discussed the activities
that characterized our sample game set. Developing this activity-based model was
still a laborious and time consuming process but ultimately successful in providing
us with a whole set of word patterns that defi ned the way professional reviewers
talk about gameplay activity. But why do we use professional game reviews?
Professional reviewers have to describe to a potential player of the game what
it is like to play, by describing the particular activities the player must undertake.
Of course that is not all they talk about. Descriptions of such things as graphics and
style, comparisons with existing games, and so on are also necessary. But profes-
sional reviewers have to discuss the game as a whole and not just their opinions of
the best and/or worst parts of a game. There are literally hundreds of thousands of
such game reviews on the World Wide Web. In the vast majority of cases, each
individual game will have multiple reviews on a variety of sites across the web.
So at this stage we had a set of activity keywords and phrases that each described
a particular gameplay activity group, and from a wide range of genres. By automati-
cally searching for these keywords and phrases within online reviews, we could use
this model to characterize games by their activity groups, not just their given genre.
The next task was to develop a prototype software system that could perform this
analysis automatically to see the kind of activity group (AG) characteristic data it
would generate. We ran the software on the same game reviews, checked the accu-
racy of the results, refi ned the model, added new activity keywords, phrases, and
AG defi nitions, and then repeated the process again and again until we were happy
with the results.
As a result of this refi nement we eventually arrived at 49 AGs, comprised of a
multitude of keywords and phrases, which in various combinations characterized
any game we could throw at the model. These can be found in Table 3.2. We were
then ready to run the analysis software on a larger set of data: as much data as we
could possibly fi nd. So we wrote software to collect URLs for as many online pro-
fessional game reviews as it could possibly deal with; hundreds of thousands of
reviews for tens of thousands of games.
Now we could run our game analysis software on all of these professional
reviews, with each game's review linked by the URLs collected. We fed the results
of this analysis into our database, which gave us a gameplay activity profi le for each
game we had found reviews for. These activity profi les consist of a set of AGs that
reviewers found more or less important for each game. Although they will not have
mentioned the activity group itself, the activity keywords and phrases that defi ne
each AG will have been mentioned numerous times in order to get a positive result.
The software calculates how importantly each AG is regarded by counting the total
number of any words in all the reviews for each game analyzed and compares this
to the number of references found for each AG. These two numbers give us a
measure of the relative importance of each activity within the game.