User Identification based on Game-Play Activity Patterns

622 views

Published on

Published in: Business, Technology
0 Comments
0 Likes
Statistics
Notes
  • Be the first to comment

  • Be the first to like this

No Downloads
Views
Total views
622
On SlideShare
0
From Embeds
0
Number of Embeds
77
Actions
Shares
0
Downloads
11
Comments
0
Likes
0
Embeds 0
No embeds

No notes for slide

User Identification based on Game-Play Activity Patterns

  1. 1. May 31, 2007 Network Game Design: User Identification based on Game-Play Activity Patterns Chun-Yang Chen, Academia Sinica Li-Wen Hong, Academia Sinica ACM NetGames 2007
  2. 2. Motivation Password-based User Identity Vulnerability Account hijacking (Identity Theft) Severity & prevalence No general solution until the victim appears Account sharing Increase the difficulty of demographical studies of game Authors / Paper Title 2
  3. 3. User Identity: Current Solutions Digital signature Smart card Biometrical signature fingerprint voice keystroke Authors / Paper Title 3
  4. 4. Our Solution A novel biometric: Game-Play Activity Patterns Authors / Paper Title 4
  5. 5. Outline Motivation Data Collection Player Activity Analysis Proposed Scheme Performance Evaluation Contribution & Future Work Authors / Paper Title 5
  6. 6. Observation More Regular More Unpredictable Motivation Data Collection Player Activity Analysis Proposed Schemes … Authors / Paper Title 6
  7. 7. Data Collection A MMORPG -- Angel’s Love A commercial game in Taiwan 40 thousands of players online The player activity logs we use Trace period of 3 days 287 randomly chosen accounts Remove logs shorter than 200 minutes Motivation Data Collection Player Activity Analysis Proposed Schemes … Authors / Paper Title 7
  8. 8. Definitions Active period An active period of a game character is defined as a time interval (t1 , t 2 ) which the character continuously moves, in with a tolerance of discontinuity up to 1 second. Idle period An idle period of a game character is defined as a time interval (t1 , t 2 ) which the character has no movements, in where t 2 − t1 ≥ second. 1 Data Collection Player Activity Analysis Proposed Schemes Performance Evaluation … Authors / Paper Title 8
  9. 9. Data Summary Player # : Data Activity Active Idle 287 Length Rate Period Period 0.35 5% 7 hr 3 sec 7 sec cycle/min 2.28 50% 51 hr 6 sec 18 sec cycle/min 5.12 95% 67 hr 9 sec 181 sec cycle/min Data Collection Player Activity Analysis Proposed Schemes Performance Evaluation … Authors / Paper Title 9
  10. 10. Distribution of Active / Idle Period 4 Data Collection Player Activity Analysis Proposed Schemes Performance Evaluation … Authors / Paper Title 10
  11. 11. Idle time is much more diverse than active time Data Collection Player Activity Analysis Proposed Schemes Performance Evaluation … Authors / Paper Title 11
  12. 12. Average Active Time v.s. Idle Time Players’ active/idle patterns can be very different candidate features for user identification Data Collection Player Activity Analysis Proposed Schemes Performance Evaluation … Authors / Paper Title 12
  13. 13. Why choosing idle time rather than active time Idle time distribution captures more variability Idle time process has smaller degree of auto- correlations Data Collection Player Activity Analysis Proposed Schemes Performance Evaluation … Authors / Paper Title 13
  14. 14. Idle Time Distribution of Random Players Data Collection Player Activity Analysis Proposed Schemes Performance Evaluation … Authors / Paper Title 14
  15. 15. KL Distances ITD: Idle time distribution RET: Relative Entropy Test relative entropy between two ITDs based on the KL distance KL distance: Kullback-Leibler distance P(i ) DKL ( P || Q) = ∑ P(i ) log i Q(i ) DSKL ( P || Q) = DSKL (Q || P) = DKL ( P || Q) + DKL (Q || P) Player Activity Analysis Proposed Schemes Performance Evaluation … Authors / Paper Title 15
  16. 16. KL distances of Players Player Activity Analysis Proposed Schemes Performance Evaluation … Authors / Paper Title 16
  17. 17. Identification Scheme: Consistency Test Perform consistency test KLD: distribution of KL distance 2 KLDs for each player KLDs are tested by two-sided Wilcoxon test 0.95 Authors / Paper Title 17
  18. 18. Identification Scheme: Discriminability Test Perform discriminability test KLDi,j : distribution of KL distances between ni ITDs of player i & nj ITDs of player j KLDs are tested by one-sided Wilcoxon test Authors / Paper Title 18
  19. 19. Factor Consideration Consideration: effect of the detection time & the history size Trec: how long the player history kept in database (in minutes) Tobs : the detection time once the player log in (in minutes) Proposed Schemes Performance Evaluation Contribution & Future Work Authors / Paper Title 19
  20. 20. Evaluation Result Proposed Schemes Performance Evaluation Contribution & Future Work Authors / Paper Title 20
  21. 21. Performance Evaluation Effect of Trec mean of activity cycles is one minute Trec = idle times assuming one million user accounts, Trec = 200 minutes, each idle time uses 4 bytes storage space = 800 MB Effect of Tobs assuming 10,000 players are online, Tobs = 20 minutes main memory = 0.8 MB Player Activity Analysis Proposed Schemes Performance Evaluation … Authors / Paper Title 21
  22. 22. Contribution & Future Work Contribution: Propose the RET scheme for user identification from the aspect of idle time. With a 20-minute detection time period given a 200-minute history size achieve higher than 90% accuracy. Future Plan Cut down the detection time Utilizing more aspects of game-play activities. Analyzing from the way users control the character. Proposed Schemes Performance Evaluation Contribution & Future Work Authors / Paper Title 22
  23. 23. Thank you! Chun-Yang Chen Li-Wen Hong ACM NetGames 2007

×