Antisocial Computing - Speaker Deck

Slide 1

Slide 1 text

Justin Cheng / @jcccf / clr3.com with Jure Leskovec, Michael Bernstein, Cristian Danescu-Niculescu-Mizil, and Tim Hsieh ANTISOCIAL COMPUTING Explaining, Predicting, and Mediating Online Negative Behavior v1.1

Slide 2

Slide 2 text

Vieweg, et al. (2010); Kittur, et al. (2013); Burke & Kraut (2016)

Slide 3

Slide 3 text

No content

Slide 4

Slide 4 text

47% of online users have been harassed Data & Society (2017)

Slide 5

Slide 5 text

Popular Science (2013); The Verge (2015); Chicago Sun-Times (2014)

Slide 6

Slide 6 text

Why is bad behavior so prevalent?  (╯°□°)╯︵ ┻━┻ Research Question

Slide 7

Slide 7 text

Understanding bad behavior helps us build healthier communities Implications Systems Guidelines Interventions

Slide 8

Slide 8 text

Antisocial behavior is largely due to sociopaths Prior Work Donath (1999); Hardaker (2010); Buckels, et al. (2014)

Slide 9

Slide 9 text

Antisocial behavior is largely due to ordinary people This Work

Slide 10

Slide 10 text

Antisocial Computing Talk Outline What causes antisocial behavior? How do systems mediate it? Does it worsen over time? 1 2 3 WIP!

Slide 11

Slide 11 text

Antisocial Computing Talk Outline What causes antisocial behavior? How do systems mediate it? Does it worsen over time? 1 2 3 WIP!

Slide 12

Slide 12 text

CSCW 2017 (Best Paper); ICWSM 2015 (Honorable Mention) with M. Bernstein, C. Danescu-Niculescu-Mizil, J. Leskovec CAN ANYONE BECOME A TROLL? Causes of Antisocial Behavior in Online Discussions

Slide 13

Slide 13 text

CONTENT WARNING!  This talk contains depictions of trolling that use strong language. !

Slide 14

Slide 14 text

No content

Slide 15

Slide 15 text

It also shows that Islam and Christianity teaching women to dress modest could be right afterall.

Slide 16

Slide 16 text

It also shows that Islam and Christianity teaching women to dress modest could be right afterall. Clearly that is the only logical conclusion to this article. Now if you'll excuse me, I need to iron my tarp. I have work on Monday, and I want to appear 'modest'. fail at life. go bomb yourself. Religious nut alert

Slide 17

Slide 17 text

We studied multiple large comment- based news communities. 470M posts 831M votes 76M users

Slide 18

Slide 18 text

What is trolling?

Slide 19

Slide 19 text

What is trolling? Engaging in negatively marked online behavior Taking pleasure in upsetting others Not following the rules Disrupting a group while staying undercover Donath (1999); Hardaker (2010); Kirman (2012); Schwartz (2008)

Slide 20

Slide 20 text

Trolling is behavior that occurs outside community norms. Deﬁned using community guidelines Our Deﬁnition e.g., name-calling, personal attacks, profanity, threats, hate speech, ethnically/racially offensive material

Slide 21

Slide 21 text

Are trolls just a vocal minority? Donath (1999); Hardaker (2010); Shachaf & Hara (2010); NYT (2008); Wired (2014); Vox (2014)

Slide 22

Slide 22 text

How much do trolls troll? Proportion of Banned Users 0 0.1 0.2 0.3 0.4 Proportion of Deleted Posts 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Slide 23

Slide 23 text

The distribution of trolls is bimodal Proportion of Banned Users 0 0.1 0.2 0.3 0.4 Proportion of Deleted Posts 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Slide 24

Slide 24 text

Are there two types of trolls? Proportion of Banned Users 0 0.1 0.2 0.3 0.4 Proportion of Deleted Posts 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Situational trolling? Lifelong trolling?

Slide 25

Slide 25 text

What if antisocial behavior is situational?

Slide 26

Slide 26 text

Challenge: how to show that antisocial behavior is situational? Observational data isn’t causal

Slide 27

Slide 27 text

Challenge: how to show that antisocial behavior is situational? Experiments hard to generalize

Slide 28

Slide 28 text

Simulated Discussion Experiment Large-Scale Analysis Solution: Experiment + Observational Study

Slide 29

Slide 29 text

Anyone can become a troll Our Hypothesis

Slide 30

Slide 30 text

“Broken windows” theory Zimbardo (1969); Wilson (1982)

Slide 31

Slide 31 text

Unpleasant stimuli increase aggression Jones & Bogat (1978); Rotton & Frey (1985)

Slide 32

Slide 32 text

Experiment: simulated discussion forum

Slide 33

Slide 33 text

N=667, 40% female Quiz Discussion Experimental method

Slide 34

Slide 34 text

Quiz Discussion Experimental method ×

Slide 35

Slide 35 text

Positive/Negative Mood Positive/Negative Context Experimental method × Quiz Discussion

Slide 36

Slide 36 text

Easy quiz (positive mood)

Slide 37

Slide 37 text

Difﬁcult quiz (negative mood)

Slide 38

Slide 38 text

Positive discussion context

Slide 39

Slide 39 text

Negative discussion context

Slide 40

Slide 40 text

How did trolling differ across conditions? Two expert raters labeled posts independently

Slide 41

Slide 41 text

How did trolling differ across conditions? Positive Mood Negative Mood Positive Context Negative Context % Troll Posts

Slide 42

Slide 42 text

Trolling is lowest with positive conditions… Positive Mood Negative Mood Positive Context 35% Negative Context % Troll Posts

Slide 43

Slide 43 text

…increases with either negative condition… Positive Mood Negative Mood Positive Context 35% 49% Negative Context 47% % Troll Posts

Slide 44

Slide 44 text

…and almost doubles in the worst case Positive Mood Negative Mood Positive Context 35% 49% Negative Context 47% 68% % Troll Posts (p < 0.05 using a mixed effects logistic regression model)

Slide 45

Slide 45 text

Negative affect almost triples Positive Mood Negative Mood Positive Context 1.1% 1.4% Negative Context 2.3% 2.9% % Negative Affect Words (LIWC) (p < 0.05)

Slide 46

Slide 46 text

Hilary is a solid candidate. As a woman, I appreciate that she's a woman, but it's not the only reason I think she would do well in ofﬁce. Positive Mood + Context

Slide 47

Slide 47 text

Anyone who votes for her is a complete idiot. These supporters are why this country is in such bad shape now. Uneducated people. Negative Mood + Context

Slide 48

Slide 48 text

Bad mood and negative discussion context increase trolling

Slide 49

Slide 49 text

Simulated Discussion Experiment Large-Scale Analysis of CNN.com Online Experiment + Observational Study

Slide 50

Slide 50 text

Can trolling, like mood, vary with the time of day and day of week? Replicating Mood Golder & Macy (2011) Neg. Affect Time of Day ?

Slide 51

Slide 51 text

How does trolling vary with time of day? Proportion of Flagged Posts 0.03 0.033 0.036 0.039 0.042 Time of Day 0 6 12 18 24

Slide 52

Slide 52 text

Trolling peaks in the evening… Proportion of Flagged Posts 0.03 0.033 0.036 0.039 0.042 Time of Day 0 6 12 18 24 Negative Affect (Golder & Macy)

Slide 53

Slide 53 text

…and early in the work week. Proportion of Flagged Posts 0.03 0.032 0.034 0.036 0.038 0.04 Day of Week Mon Tue Wed Thu Fri Sat Sun

Slide 54

Slide 54 text

Trolling peaks when moods are worse Time of Day Proportion of Flagged Posts Negative Affect Proportion of Downvotes Day of Week

Slide 55

Slide 55 text

Mood spills over from prior discussions Replicating Mood

Slide 56

Slide 56 text

… … (Discussion) Mood spills over from prior discussions

Slide 57

Slide 57 text

… … … ? … (Discussion) (Unrelated Discussions) … ? … Mood spills over from prior discussions

Slide 58

Slide 58 text

… … … … (Discussion) … … Mood spills over from prior discussions

Slide 59

Slide 59 text

A user who trolled in a previous discussion is twice as likely to troll in a later, unrelated discussion (p < 0.01) Replicating Mood

Slide 60

Slide 60 text

The initial post affects subsequent trolling Replicating Context

Slide 61

Slide 61 text

(Separate discussions of same article) … ? ? ? … ? ? ? The initial post affects subsequent trolling

Slide 62

Slide 62 text

(Separate discussions of same article) … … The initial post affects subsequent trolling

Slide 63

Slide 63 text

An initial troll post increases the subsequent trolling by 63% (p < 0.01) Replicating Context

Slide 64

Slide 64 text

Troll or not? User

Slide 65

Slide 65 text

Troll or not? User Mood

Slide 66

Slide 66 text

Troll or not? User Mood Other users { }

Slide 67

Slide 67 text

Because trolling is situational, ordinary people can end up trolling

Slide 68

Slide 68 text

No content

Slide 69

Slide 69 text

Can voting mitigate bad behavior?

Slide 70

Slide 70 text

Downvoting causes negative behavior to worsen Our Hypothesis

Slide 71

Slide 71 text

Antisocial Computing Talk Outline What causes antisocial behavior? How do systems mediate it? Does it worsen over time? 1 2 3 WIP!

Slide 72

Slide 72 text

How Antisocial Behavior Worsens ICWSM 2014 with C. Danescu-Niculescu-Mizil, J. Leskovec CAN ANTISOCIAL BEHAVIOR SPI R A L?

Slide 73

Slide 73 text

Downvoting causes negative behavior to worsen Our Hypothesis

Slide 74

Slide 74 text

What effects do evaluations have? Positively evaluated ? ? Negatively evaluated

Slide 75

Slide 75 text

What is a positive or negative evaluation?

Slide 76

Slide 76 text

Deﬁning positive and negative evaluations : 9 1 2 8 N↑  N↑ + N↓ 2  2+8 = = 0.2 p↓ ≤ : N↑  N↑ + N↓ 9  9+1 = = 0.9 p↑ ≥ Positive Evaluation Negative Evaluation (validated using a crowdsourcing experiment)

Slide 77

Slide 77 text

What effects do evaluations have? Positively evaluated ? ? Negatively evaluated

Slide 78

Slide 78 text

Does feedback encourage better behavior? Skinner (1938)

Slide 79

Slide 79 text

Or is bad stronger than good? Brinko (1993); Baumeister, et al. (2001)

Slide 80

Slide 80 text

Four large comment-based news communities

Slide 81

Slide 81 text

What effects do evaluations have? Positively evaluated Negatively evaluated

Slide 82

Slide 82 text

What effects do evaluations have? … … … …

Slide 83

Slide 83 text

What effects do evaluations have? … … … … Before After vs. Before After vs.

Slide 84

Slide 84 text

Challenge: how to compare different users and posts? Aren’t downvoted users/posts inherently worse?

Slide 85

Slide 85 text

Solution: propensity score matching PSM: Rosenbaum (1983); CEM: Iacus, et al. (2012) Positively evaluated Negatively evaluated

Slide 86

Slide 86 text

Match on text quality Similar text quality q } ≈

Slide 87

Slide 87 text

Computing text quality (Validated using crowdsourcing) Learn p with bigrams 1 3 lorem ipsum… q = ? Lorem… ? ? 9 2 lorem ipsum… … Text quality q is predicted p

Slide 88

Slide 88 text

Match on text quality Similar text quality q(c↑ )=q(c↓ ) } ≈

Slide 89

Slide 89 text

…as well as other covariates Similar history (# posts, overall proportion of upvotes, etc.) { ≈ … … ≈

Slide 90

Slide 90 text

…as well as other covariates ≈ … … … … ≈

Slide 91

Slide 91 text

How are subsequent posts evaluated? ≈ … … … … ≈

Slide 92

Slide 92 text

How much are evaluations due to textual or community effects?

Slide 93

Slide 93 text

How much are evaluations due to textual effects (i.e., people writing worse)? f***ing a****** i.e., downvoting because of post content

Slide 94

Slide 94 text

How much are evaluations due to community effects (i.e., inherent bias)? We dislike you. i.e., downvoting because of community dislikes author We dislike you.

Slide 95

Slide 95 text

Do people write better/worse after a positive/negative evaluation? Textual Effects

Slide 96

Slide 96 text

≈ … … … … Better/Worse? Do people write better/worse after a positive/negative evaluation? ≈

Slide 97

Slide 97 text

Text quality drops signiﬁcantly after a negative evaluation… (p < 0.05, mean effect size r = 0.18) … … Negativity bias

Slide 98

Slide 98 text

…but doesn’t change after a positive evaluation … … (n.s.) Negativity bias

Slide 99

Slide 99 text

How does community bias change after an evaluation? Community Effects

Slide 100

Slide 100 text

… … … … How does community bias change after an evaluation?

Slide 101

Slide 101 text

Measuring community bias N↑ q N↓ N↑  N↑ + N↓ = 0.5 p(c) = = 0.8 q(c) p(c) q(c) = 0.3 Prop. Upvotes Text Quality Community Bias − −

Slide 102

Slide 102 text

Community bias increase more after a negative than positive evaluation (p < 0.01, mean effect size r = 0.13) … … Halo effect

Slide 103

Slide 103 text

Troll or not? User Mood Other users { }

Slide 104

Slide 104 text

Troll or not? User Mood Other users { }

Slide 105

Slide 105 text

Troll or not? User Mood Other users { }

Slide 106

Slide 106 text

Trolls may start out normal, but tip into a spiral and never recover

Slide 107

Slide 107 text

Were downvotes a good idea in the ﬁrst place?

Slide 108

Slide 108 text

Antisocial Computing Talk Outline What causes antisocial behavior? How do systems mediate it? Does it worsen over time? 1 2 3 WIP!

Slide 109

Slide 109 text

Work-in-Progress with J. Leskovec, T. Hsieh WIP! How Introducing Downvoting to Communities Impacts User Behavior CAN ENABLING NEGATIVE FEEDBACK HARM COMMUNITIES?

Slide 110

Slide 110 text

Downvoting causes negative behavior to worsen… Previously WIP!

Slide 111

Slide 111 text

…but does its mere presence affect communities? The Present Work WIP!

Slide 112

Slide 112 text

The ability to downvote negatively alters behavior Our Hypothesis WIP!

Slide 113

Slide 113 text

Challenge: how do we measure the impact of introducing downvoting? WIP!

Slide 114

Slide 114 text

Why the sudden increase in downvotes? Data: Worldstarhiphop circa. 2012 # Votes 0 40000 80000 120000 160000 Time (Days) 0 20 40 60 80 100 120 140 160 180 200 220 Downvotes Upvotes ? WIP!

Slide 115

Slide 115 text

Early 2012 Late 2012 Cause: a Disqus interface change WIP!

Slide 116

Slide 116 text

Early 2012 Late 2012 Cause: a Disqus interface change + WIP!

Slide 117

Slide 117 text

A natural experiment studying the impact of introducing downvoting Opportunity

Slide 118

Slide 118 text

Twenty comment-based websites … … WIP!

Slide 119

Slide 119 text

How do we measure impact? WIP! Time (Weeks) 0

Slide 120

Slide 120 text

How do we measure impact? WIP! Time (Weeks) 1 0

Slide 121

Slide 121 text

How do we measure impact? WIP! Time (Weeks) 1 5 -4 0 Before After Repeat for each domain (x20), applying corrections as necessary

Slide 122

Slide 122 text

How did the introduction of downvoting affect user behavior? Votes Discussions Anonymity WIP!

Slide 123

Slide 123 text

Does introducing downvoting alter upvoting behavior? WIP!

Slide 124

Slide 124 text

Does introducing downvoting alter upvoting behavior? ? ? Downvotes don’t impact upvotes overall Upvotes increase to offset downvotes WIP! (Muchnik, et al. 2013)

Slide 125

Slide 125 text

Introducing downvoting has little effect on upvoting overall. (5 domains saw a signiﬁcant increase / 12 saw no change / 3 saw a signiﬁcant decrease, overall n.s.) WIP!

Slide 126

Slide 126 text

Does introducing downvoting spur discussion? WIP!

Slide 127

Slide 127 text

Does introducing downvoting spur discussion? ? ? Downvotes cause arguments Downvotes replace comments WIP!

Slide 128

Slide 128 text

Introducing downvoting spurs discussion. More replies, less top-level comments (13 + / 4 · / 3 —, overall increase [p < 0.01]) WIP!

Slide 129

Slide 129 text

Introducing downvoting spurs discussion. More replies, less top-level comments Discussion length increases … (13 + / 4 · / 3 —, overall increase [p < 0.01]) (15 + / 3 · / 2 —, overall increase [p < 0.01]) WIP!

Slide 130

Slide 130 text

Introducing downvoting spurs discussion. More replies, less top-level comments Discussion length increases More back-and-forth discussion … @john… @mary… @john… @mary… (13 + / 4 · / 3 —, overall increase [p < 0.01]) (15 + / 3 · / 2 —, overall increase [p < 0.01]) (16 + / 2 · / 2 —, overall increase [p < 0.01]) WIP!

Slide 131

Slide 131 text

Does introducing downvoting have a “chilling effect”? WIP!

Slide 132

Slide 132 text

Introducing downvoting increases anonymity. WIP!

Slide 133

Slide 133 text

Introducing downvoting increases anonymity. WIP! (18 sig. increase / 1 no change / 1 sig. decrease, overall increase [p < 0.01])

Slide 134

Slide 134 text

Negative feedback mechanisms can (unintentionally) encourage trolling

Slide 135

Slide 135 text

What we now know What we thought Trolls are a vocal minority Trolls can be ordinary people Trolling is innate Trolling can spiral from a single bad post Design mediates trolling behavior Design mediates a user’s experience ANTISOCIAL COMPUTING