Detecting Bad Mouthing Behavior in Reputation Systems
Upcoming SlideShare
Loading in...5
×
 

Detecting Bad Mouthing Behavior in Reputation Systems

on

  • 866 views

 

Statistics

Views

Total Views
866
Views on SlideShare
855
Embed Views
11

Actions

Likes
0
Downloads
9
Comments
0

2 Embeds 11

http://mmnet.iis.sinica.edu.tw 9
http://www.slideshare.net 2

Accessibility

Upload Details

Uploaded via as Adobe PDF

Usage Rights

© All Rights Reserved

Report content

Flagged as inappropriate Flag as inappropriate
Flag as inappropriate

Select your reason for flagging this presentation as inappropriate.

Cancel
  • Full Name Full Name Comment goes here.
    Are you sure you want to
    Your message goes here
    Processing…
Post Comment
Edit your comment

Detecting Bad Mouthing Behavior in Reputation Systems Detecting Bad Mouthing Behavior in Reputation Systems Presentation Transcript

  • Detecting Bad Mouthing Behavior in Reputation Systems Kuan-Ta Chen (Chun-Yang Chen) , Cheng-Chun Lou , Polly Huang , and Ling-Jyh Chen 1 2 2 1 1 Academia Sinica, 2National Taiwan University Background Hypothesis Votee-Voter-based Collusion Cluster Detection • MMORPGs (Massively Multiplayer Online Role-Playing Games) • The most voters of legitimate players are likely collusion cluster The relationship of victim group is stronger than random votee cluster have become extremely popular • In case if a collusion cluster attacks for several times First take the votees with more common voters 1. Collusion Cluster has more common votees than random • form the victim group • Game bots voter cluster ○ Auto-playing game clients 2. Victims have more common voters than random votee cluster Then take union of the voters of the victim group ○ One of the greatest threats of MMORPGs • form the collusion cluster candidate • Based-on Apply Voter-based Collusion Cluster Detection • Detection of Game Bots ○ The voters & votees id of each player ○ Manual detection (game master) [1] • Note ○ Traffic analysis approach [2] ○ When not attack, a collusive player acts as a legitimate player Performance Evaluation Results ○ Voting-based system [3] Voter-Based Votee-Voter-Based Comparison - Each player votes the suspicious player as a game bot Voter-Based Collusion Cluster Detection # of Attack (Single CC or Multiple CC) More attacks  higher accuracy Votee-Voter > Voter The relationship of collusion cluster is stronger than random voter cluster No influence Need more than 3 Motivation Take the voters with more common votees between each other Collusion Cluster Size high accuracy high accuracy Voter > Votee-Voter Prob. of Collusive Player • form the collusion cluster. Attack Vote Higher Prob.  higher accuracy Votee-Voter > Voter • Problem in voting-based system General Case Malicious Case 3 ID Votee List 4 1 1 11, 12, 13, 14, 17 Voter-Based scheme: 5 4 2 2 11, 12, 13, 14, 15, 18 Robust to the size of collusion cluster 1 4 3 9, 13, 14, 15 4 3 3 4 11, 12, 13, 14 Voter-Votee-Based scheme: BOT BOT 1 Robust to the number of attacks BO BO BO BO 2 4 1 5 12, 13, 14, 15, 25, 60 1 T T T T 2 1 9 17, 25, 30 BO T B OT 1 3 3 18 10 17, 18, 50, 56 Conclusion 10 11 17, 18, 70, 44 • Two mechanisms to detect the collusion cluster BOT User 1 1 12 17, 20, 35 ○ Based on the voting history 1 • Collusion 13 13 40, 45, 50 ○ Single cluster or multiple clusters 1 14 55, 80, 90 ○ A secret agreement between two or more parties for a fraudulent, 17 9 1 illegal, or deceitful purpose [4] 1 1 15 20, 30, 40 • Accuracy - Unfairly low ratings – bad mouthing 14 1 1 17 1 ○ Attack more than three times: 83%+ 1 - Unfairly high ratings – ballot stuffing 18 17, 60 ○ Attack more than five times: 97%+ 15 1 12 11 Weight (A, B): # of common • Only can vote negatively (game bot) 1 votees of A and B • Adjust other experimental factors ○ Only collusion cluster size and prob. of collusive player attack vote ○ This study focuses on bad-mouthing attacks < Algorithm > have the obvious influence to the accuracy while ( edge(G ) ∉ Φ ) Problem Formulation Take the edge with the largest weight • Bad-Mouthing while (not termination condition ) Furture Work Take the outlier edge with the largest weight Detecting players who participate in multiple collusion clusters ○ A malicious group deliberately vote a legitimate player as a end while game bot output [1] I. MacInnes and L. Hu, “Business models and operational issues in the chinese online game • Terms end while industry,” Telematics and Informatics, vol. 24, no. 2, pp. 130-144, 2007 ○ Collusion Cluster: a bad-mouthing group [2] K.-T. Chen, J.-W. Jiang, P. Huang, H.-H. Chu, C.-L. Lei, and W.-C. Chen, “Identifying mmor- Terminal Condition: ○ Victim: the legitimate players who are under bad-mouthing attacks ∑{weight (u,v) | u, v ⊂ S ' } − ∑{weight (u,v) | u, v ⊂ S} < a pg bots: A traffic analysis approach,” in Proceedings of ACM SIGCHI ACE, 2006 • Goal S [3] Blizzard, http://www.blizzard.com/war3/ To Detect the Collusion Clusters [4] http://www.answers.com/collusion?cat=biz-fin