background image
1.2
Terminology
In order to have a clear vocabulary to describe
data mining in multiplayer games (from now
on called game mining) we define three main
types:
1) game content mining ­ discovery
of patterns in multimedia or textual content
in games (e.g. room layout), 2) game struc-
ture mining ­ discovery of structural patterns
in form of paths and connections binding the
game world together (e.g. hallways between
rooms), and 3) game usage mining ­ discovery
of human and avatar behavior patterns. The
described types of game mining are inspired by
well-known types of web mining ­ web content
mining, web structure mining and web usage
mining [
2
].
1.3
Research Problem
What are the similarities and differences of in-
formation gathering for web usage mining and
game usage mining?
The rest of this paper is organized as fol-
lows.
Section 2 describes a game classifica-
tion schema. Section 3 describes the Game Us-
age Mining concept. Section 4 compares infor-
mation gathering in a web and game context.
Section 5 describes the proposed game log ap-
proach, and finally the conclusion with future
work.
2
Game Classification
From
a
information
gathering
viewpoint,
games are proposed classified according to fig-
ure
1
, with examples for each class.
Singleplayer
(SPG)
Multiplayer
(MPG)
Massive Multiplayer
(MMPG)
Discrete
Gamestate
Non-Discrete
Gamestate
Solitaire, Tetris
Go, Backgammon,
Chinese Checkers
?
Quake,
MUD/MOO
Everquest,
Lineage
Zelda, The
Longest Journey
Figure 1: Computer Game Classification
In the discrete game-state class we consider
games that have a discrete search space and a
turn-based gameplay, e.g. in chinese checkers
it is not possible to do fractional (non-integer)
moves of marbles, and players have to wait
until their turn.
In Quake [
1
] the search
space (from a practical view) is non-discrete,
and players don't have to wait until their
turn, hence Quake belongs to the non-discrete
game-state class.
In the singleplayer game class, the games
only have one simultaneous (human) player,
as in case of Solitaire where the human plays,
and the computer only shuffles the cards.
The difference between the multiplayer and
the massive multiplayer classes is not absolute,
but the prior covers games with 2 - 100 simul-
taneous players (or same order of magnitude),
and the latter covers games with the number
of players being
100.
An example of a
(non-discrete game-state) massive multiplayer
game is NCSoft's Lineage with approximately
110,000 simultaneous players [
4
]. We were not
able to find examples of discrete gamestate
massive multiplayer games.
Another classification schema for computer
games is based on genres (e.g. action games,
role-playing games, adventure games, sports
game etc.) [
6
]. Genres are suited to classify
what the game is about, but not so well suited
to classify the amount and what type of data
that can be gathered from the game, hence
being less useful in a data mining context.
Games can also be classified according to
their network architecture (e.g. single node,
peer-to-peer, client/server or server-network)
[
8
], which is useful for describing where to col-
lect the data, but doesn't say anything about
what type of data to collect.
Paper E
101

<< - < - > - >>