Download - BRANDED WITH A SCARLET “C”: CHEATERS IN A … WITH A SCARLET “C”: CHEATERS IN A GAMING ... • Nickname (mutable) • Date ... Social ties as predictor of cheating 27

BRANDED WITH A SCARLET “C”: CHEATERS IN A GAMING SOCIAL NETWORK Jeremy Blackburn, Ramanuja Simha, Nicolas Kourtelis, Xiang Zhou, Matei Ripeanu, John Skvoretz, and Adriana Iamnitchi University of South Florida University of British Columbia

1

Video games are a huge industry

• Modern Warfare 2 released Nov. 2009 •  First 24 hours of release

•  4.7 million units sold •  $310 million in revenue

•  First 5 days of release •  8 million online players

• All these numbers eclipsed by MW3 in 2011!

2

Biggest entertainment

launch in history!

Multiplayer gaming: growing eSports industry

Major League Gaming claims 225% growth from 2010 to 2011

3

“Flash” makes $250k a year

playing StarCraft!

Team Na’Vi won $1 million in the DOTA Intl.

Tournament

But not all is well…

•  Fame and fortune attracts deviant behavior • Virtual goods worth $ attract criminal element • Competitive gameplay attracts cheaters

•  Multiplayer games are a distributed system •  Some computation left to gamers’ machines •  Susceptible to attacks

•  $100k a year to cheat creators for single game

4

Real world cheat: Wallhack

5

Players should not be visible (they are behind the wall).

What can we learn from a gaming community? • Social systems have unethical actors • Cheating in games is black and white •  Theories indicate unethical behavior has a social

component

6

What are the network characteristics of unethical actors in a large scale online community?

Steam Community

•  Large online social network for PC gamers • Built on top of Steam digital delivery platform • Purchased games permanently tied to account • Steam account required to create Steam Community

profile •  Steam Community profile not required to play games

7

Steam Community Profile • Unique SteamID •  Friends list • User specified location • Cheating flag (VAC ban) • Nickname (mutable) • Date of account creation • Screenshots • Videos • Comments (“wall posts”) • Profile information • Game reviews • Gameplay ownership/stats • Virtual goods inventory

8

Cheating flag

The cheating flag

•  Cheating automatically detected via Valve Anti Cheat system •  Method and timestamp not public •  Delayed application

•  Permanent •  Publicly viewable

•  Even private accounts

•  Can’t play on VAC secured servers •  Only applies to the game that was cheated in

•  Most servers are VAC secured •  4,200 of 4,234 Team Fortress 2 servers

•  Cheater not permanently removed from Steam Community

9

Steam Community data set

• Data collected March 16 – April 3, 2011 •  Distributed BFS using Amazon EC2

• Cheaters make up 7% of profiles •  7% of cheaters have private profiles

•  3% of non-cheaters with private profiles

• Cheaters as likely to be friends-only as private •  Non-cheaters about 3 times as likely to be friends-only as private

10

Cheaters more likely to be

private

Type Nodes Edges Profiles Public Private Friends-only

Location set

All users 12,479,765 88,557,725 10,191,296 9,025,656 313,710 851,930 4,681,829

Cheaters - - 720,469 628,025 46,270 46,714 312,354

Cheaters more likely to be private than non-cheaters

• How are cheaters positioned? •  In the social community •  Geographically

• What is the reaction to the cheating brand? •  From cheaters themselves •  In the social network •  In game

• Does the social structure influence cheating?

11

Observing the gaming community




12


13

CC

DF:

P(d

egre

e ≥

x)

10-8

10-7

10-6

10-5

10-4

10-3

10-2

10-1

100

100 101 102 103

Degree

Steam CommunityCheaters Only

Cheaters are well embedded…

…but are not central

• Cheaters under-represented among most central players •  Cheaters make up 7% of player population, but far less than 7% of

the top 0.1% central users •  Not adequately represented until top 5% central users

14

Top-N% 0.1 0.5 1.0 5.0 10.0 Degree 3.25 4.46 5.11 7.06 8.20

Betweenness 5.16 5.95 6.35 7.86 8.58

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1Fraction of cheater friends

CheatersNon-cheaters

Cheaters have more cheater friends

15

CD

F: P

(frac

tion ≤

x)

70% of cheaters’ friends lists are at least 10% cheaters

15% of cheaters have mostly cheater friends

Fraction of cheaters in neighborhood

0.000.020.040.060.080.100.120.140.160.18

USABrazil

RussiaGermany

France

UK Poland

Canada

Australia

Sweden

Denmark

Norway

Rat

io

Cheater : Non-cheater

Non-uniform geo-political distribution

16

Ratio of cheaters to non-cheaters

Cheaters are geographically closer

17

Network # of nodes # of edges <Duv> (km) <luv> (km) <NL> Steam Community 4,342,670 26,475,896 5,896 1,853 0.79 Cheater-to-Cheater 190,041 353,331 4,607 1,761 0.79 BrightKite 54,190 213,668 5,683 2,041 0.82 FourSquare 58,424 351,216 4,312 1,296 0.85

0

0.2

0.4

0.6

0.8

1

0 0.2 0.4 0.6 0.8 1Node Locality

CDF

Steam CommunityCheater-to-Cheater

Cheaters Only

CD

F: P

(nod

e lo

calit

y ≤

x)




18


Cheaters try to hide when caught…

• Recrawl in October, 2011 •  43,465 non-cheaters now flagged as cheaters •  13% had privacy setting change

•  Compared to a bit more than 3% of non-cheaters

•  10% from public to more restrictive setting •  Compared to less than 3% of non-cheaters

19

…and for good reason: the community disapproves

Change in Degree

Cheaters Non-cheaters

Net loss 44% 25%

Net gain 13% 36%

No change 43% 39%

20

Cheaters tend to lose friends while non-cheaters tend to gain friends

Gameplay logs

• Team-based, objective oriented •  Two teams, nine classes •  “Friend” interactions •  “Foe” interactions

• Popular TF2 server •  VAC secured •  Community owned •  April 1 - June 8, 2011

•  Interaction network

•  10,354 players •  93 cheaters •  486,808 edges

21

Cheaters not mistreated in games

22

10-5

10-4

10-3

10-2

10-1

100

100 101 102 103 104

Unique gameplay friends

CC

DF

Non-cheatersCheaters

10-5

10-4

10-3

10-2

10-1

100

100 101 102 103 104

Unique gameplay foes

Non-cheatersCheaters

CC

DF:

P(in

tera

ctio

n pa

rtner

s ≥

x)

Number of distinct interaction “friends” Number of distinct interaction “foes”

CC

DF:

P(in

tera

ctio

n pa

rtner

s ≥

x)




23


Does cheating spread over social links?

•  Label nodes with the date of their VAC ban •  180-day snapshots of the cheater status of nodes over

time •  For each snapshot, only those players whose ban date is from a

previous snapshot are treated as cheaters

24

Do the neighborhoods for newly-marked cheaters differ from those of non-cheaters?

Historical ban dates • 3rd party web site, vacbanned.com, provides historical data on when a VAC ban was first observed • Dates must be treated as banned “on or before”

25

0 0.2 0.4 0.6 0.8

1

2009-11-01

2010-01-01

2010-03-01

2010-05-01

2010-07-01

2010-09-01

2010-11-01

2011-01-01

2011-03-01

2011-05-01

2011-07-01

2011-09-01

2011-11-01

CD

F

Attempt made to populate database by vacbanned.com administrators in May, 2011

P(b

an o

bser

ved

befo

re d

ate)

26

10-6

10-5

10-4

10-3

10-2

10-1

100

100 101 102

Number of cheater friends

Interval starting 2009-12-30

CC

DF


10-6

10-5

10-4

10-3

10-2

10-1

100

100 101 102



10-6

10-5

10-4

10-3

10-2

10-1

100

100 101 102



10-6

10-5

10-4

10-3

10-2

10-1

100

100 101 102



10-610-410-2

101 102

0

0.2

0.4

0.6

0.8

1

0 0.2 0.4 0.6 0.8 1Fraction of cheater friends


CD

F


0

0.2

0.4

0.6

0.8

1




0

0.2

0.4

0.6

0.8

1




0

0.2

0.4

0.6

0.8

1




0.4

0.8

0.4 0.8

CC

DF:

P(n

um c

heat

er fr

iend

s ≥

x)

CD

F: P

(frac

che

ater

frie

nds ≤

x)

Evolution of cheaters’ social structure

100

101

102

103

104

105

106

107

10 20 30 40 50 60 70 0

0.02

0.04

0.06

0.08

0.1

0.12

0.14

0.16

Num

ber o

f pla

yers

Prob

abilit

y


New cheatersNon-cheaters

Probability to cheatSmoothed probablity

Social ties as predictor of cheating 27

(*plot not in paper)

Num

ber o

f pla

yers

Pch

eat(n

um c

heat

er fr

iend

s)

•  Increasing probability of a player becoming a cheater as the number of cheaters in his social neighborhood increases*

•  Decision tree classifier had ROCA of 0.61 based on number of cheater friends

Summary of results

• Homophily between cheaters •  Even though cooperation not necessary

• Cheaters’ distribution not uniform •  In social network •  Geo-politically

• Cheaters face social penalty •  But not in game

• Cheating behavior spreads via social links •  Number of cheater friends predictor of future cheating

28

Impact

•  Large scale study of unethical actors in online community •  Correlation of unethical behavior and network structure •  Useful for building models of unethical behavior

• Cheating is a social problem •  Community serves out social punishment •  Suggests exploring other social solutions for deviant behavior

• Scale of cheating of particular concern for gamified systems •  Our study exposes a likely lower bound on cheating behavior •  Social predictors can narrow focus to at-risk cheaters

29

30

Ranked locations of cheaters (red = more)

BACKUP SLIDES

31

Social Network Data Collection

• Distributed BFS •  6,445 random seeds • Up to 6 `m1.medium’

instances • Coordinated with

Amazon SQS queue •  Two crawls

•  March 16th – April 3rd, 2011 •  October 18th – October 29th, 2011

32

Internet

steamcommunity.com

CrawlerCrawler

Results Database

CrawlerCrawler

Crawler

ID2ID3ID4

SQS Queue

ID1

• More interaction partners (even on a single server) than declared Steam friends

• Steam Community friendships have meaning in-game •  Declared friends likely to have more interactions with each other

than non-friends

33

10-810-710-610-510-410-310-210-1100

100 101 102 103 104

Degree

CCDF

SocialInteraction

10-810-710-610-510-410-310-210-1100

100 101 102 103 104

Number of interactions per pair

Non-declaredDeclaredC

CD

F: P

(deg

ree ≥

x)

CC

DF:

P(n

um in

tera

ctio

ns ≥

x)

Interaction vs. declared relationships

0.7

0.75

0.8

0.85

0.9

0.95

1

100 101 102 103

Net change in neighborhood size

CDF

Cheaters NegativeCheaters Positive

Non-cheaters NegativeNon-cheaters Positive

Change in Degree Cheaters Non-cheaters

No change 43% 39% Net gain 13% 36% Net loss 44% 25%

34

Cheaters tend to lose friends while non-cheaters tend to gain them Cheaters lost 2x the number of friends that they gained

CD

F: P

(cha

nge

in d

egre

e ≤

x) The community is disapproving

• Overlap between two neighborhoods is an indicator of social strength

• Cheaters form stronger relationships with each other than with non-cheaters •  Overlap of cheater-cheater and non-cheater-non-cheater

neighborhoods greater than overlap of mixed neighborhoods

CD

F: P

(ove

rlap ≤

x)

Do cheaters form strong relationships?

35

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9

1

0.001 0.01 0.1 1Overlap in friendships network

CDF

CheatersNon-cheatersCheaters, C-CNon-cheaters, NC-NC

Distribution of cheater friends

• Cheaters likely to have more cheater friends when compared to non-cheaters

36

10-810-710-610-510-410-310-210-1100

100 101 102 103

Cheater Degree

CC

DF

CC

DF:

P(d

egre

e ≥

x)

Privacy settings and inferred degree

• Position in network inferable from your public friends

37

10-810-710-610-510-410-310-210-1100

100 101 102 103

Degree

No ProfilePublicFriends-OnlyPrivate

Real world cheat: Wallhack

38

Player partially obscured, hiding

behind cover

Cheating behavior outside of gaming • Academics

• Plagiarism • Cheating on exams

• Personal relationships • Cheating on your spouse

• Real-world crime • Confidence scams

• Minor law breaking • Speeding

• Real-world sports • Steroids •  “Diving” in soccer

• Cyber crime • Spam • Malware distribution

• Business • Accounting fraud • Corporate espionage

39