examining game world topology personalization
TRANSCRIPT
Examining Game World Topology Personalization
Sauvik Das
Carnegie Mellon University
Alexander Zook
Georgia Institute of Technology
Mark Riedl
Georgia Institute of Technology
1
2
Is it too linear?
3
4
5
“Boring. Bland. Linear. Those are three of the kinder things that can be said of Final Fantasy XIII…”
http://kotaku.com/5881121/just-why-exactly-did-final-fantasy-xiii-suck-so-badly
6
http://www.gameinformer.com/blogs/editors/b/gijoe_blog/archive/2010/03/09/yes-crybaby-final-fantasy-xiii-is-linear.aspx
“You want to make steady progress without getting bogged down by ancillary garbage? Heresy!”
7
Players care about topology
• Final Fantasy XIII is just an example — illustrative of a broader point.
• Anecdotally, players have strong personal preferences for game world topology.
8
What if each player got their desired game world?
9
Our contribution
• In that spirit, our goal was to examine whether:
(i) personalizing game world topologies increases user engagement with a Computer Role-Playing Game (i.e., is there value in personalizing topologies?)
(ii) player preferences for game world topology can be predicted (i.e., can we algorithmically determine how to best personalize topologies?)
10
Method
11
The Game• Turn-based combat.
• Regular and rare monsters.
• Boss battles.
• Treasure chests in sidepaths.
• Story: simplified Yuffie Wutai sidequest from FF7.
12
“Big” world “Small” world
13
Within-subjects study
• We ran a within-subjects study with 16 participants.
• Participants played a game three times. The only difference between each play through was the topology of the game.
• First world: baseline; topologically “neutral”
• Second/third worlds: “big” or “small”, randomly ordered
• Afterwards, players filled out survey providing complete ordering of world preference.
14
Data-collected
15
Results
16
Descriptive Stats
• 16 participants. 14 male. Mostly early twenties - early thirties.
• On average, 45 minutes to complete study (27-83 range).
• Strong topological preferences: Mostly preferred big (7) or small world (7). Two preferred neutral (2). Those two also preferred small to big.
• order of presentation had non-significant effect on preference.
17
Are players more engaged in topologically aligned worlds?• Operationalized engagement as higher overall activity (more of the
world explored, more battles, more treasures opened, etc.)
• Divided players into groups based on order of presentation (big first or small first) and world preference (big or small). Compared engagement within order and preference groups
• BB: Big first, big preference
• BS: Big first, small preference
• SB: Small first, big preference
• SS: Small first, small preference
18
Expectation
• Controlling for order of presentation, players should be more engaged in the world that better aligns with their topological preference.
• in big world: BB > BS and SB > SS
• in small world: BS > BB and SS > SB
19
20
Big First
0
15
30
45
60
Sidepaths Treasures Battles Time Exploring
Big Preference Small Preference
Small First
0
12.5
25
37.5
50
Sidepaths Treasures Battles Time Exploring
Big Preference Small Preference
Big World Engagement
21
Big First
0
7.5
15
22.5
30
Sidepaths Treasures Battles Time Exploring
Big Preference Small Preference
Small First
0
4
8
12
16
Sidepaths Treasures Battles Time Exploring
Big Preference Small Preference
Small World Engagement
Are players more engaged in topologically aligned worlds?
Yeah, probably.
22
Can we predict player topological preferences from in-game behavior?
• Created ridge-logistic regression model (predicting for big or small preference).
• Dataset: in-game behavioral trace from playthrough of initial baseline world.
• Used Wrapper feature subset selector to pick best 4 predictive features to avoid overfitting.
• Used ridge regularization to avoid overfitting.
23
Results
• With leave-one-out CV, correctly predicted topology preference for 13/16 participants (81%).
• 3 errors were all those who preferred large world, but predicted to prefer small.
24
Can we predict player topological preferences from in-game behavior?
For a lot of people, yes.
25
Conclusion
26
We ran a small within-subjects study, examining whether: (i) personalizing CRPG world topologies can increase
engagement with a game;
(ii) in-game behavior can predict a player’s preferences for game world topology.
“Big” world “Small” world27
Player behavior in an initially topologically neutral world can be used to predict their topological preferences.
Players may exhibit greater engagement in worlds that better align with their topological preferences.
Players have strong preferences for game world topology.
28
Thus, real-time algorithmic personalization of game world topology may be both possible and valuable for CRPGs.
29
1. Players have strong preferences for game world topologies. 2. Player behavior in an initially topologically neutral world can be
used to predict their topological preferences. 3. Players may exhibit greater engagement in worlds that better
align with their topological preferences.
Take-aways
Sauvik Das [[email protected]]
Carnegie Mellon University
Examining Game World Topology Personalization
Results
• With leave-one-out CV, correctly predicted topology preference for 13/16 participants (81%).
• 3 errors were all those who preferred large world, but predicted to prefer small.
31
GameForge
32
33