Information Diffusion 1: Social Contagion

2

Experimental evidence of massive-scale emotional contagion through social networks

Adam D. I. Kramer, Jamie E. Guillory, and Jeffrey T. Hancock

(covered by Ruiqi Zhu)

Rumor cascades

Adrien Friggeri, Lada A. Adamic, Dean Eckles, Justin Cheng

(covered by Zexing Song and Haomin Lin)

1.

Experimental evidence of massive-scale emotional contagion through social network

Adam D. I. Kramer, Jamie E. Guillory, Jeffrey T. Hancock

3

What is Contagion?

• “Transmission of a disease by a disease by directed contact with an infected person or object; a disease or poison transmitted in this way; the means of transmission; the transmission of an emotional state, e.g. excitement; a harmful influence (Microsoft dictionary 1997)

4

Social Contagion

• Human behavior clusters in both space and time even in the absence of coercion and rationale
• Emotional contagion (the spread of mood and affect through populations by simple exposure)
• Behavioral contagion (the spread of behaviors through populations by simple exposure)

5

Research on Social Contagion

• Network phenomena appear to be relevant to the behavioral trait of obesity (Christakis et al. 2007)
• Significant influence of emotional contagion on individual-level attitudes and group processes (Bardade 2002)
• Mimicry and feedback as mechanisms (Hatfield et al. 1993)
• Support for “emotional similarity hypothesis” and mimicry (Gump et al. 1997)

6

Motivation & Research Gap

1. Most studies have been correlational (misspecification of contextual variables & failure to account for shared experience)
2. Controlled experiments examine emotions after social interactions
3. Whether nonverbal cues are necessary for emotional contagion
4. “Alone together” social comparison effect leads to mixed result
5. No causal evidence of emotional contagion

7

Research Goal

• Provide causal evidence of emotional contagion in a massive online social network

​

8

Why it is important?

• Emotional contagion could affect the emotions of groups and individuals
• Social networks enable emotional contagion to occur on a massive scale
• Causal evidence of emotional contagion could help us better understand, predict, measure, and control its effect

9

Prior Works

1. Emotional contagion occurs via text-based computer- mediated communication
2. Contagion of psychological and physiological qualities has been suggested based on correlational data for social networks generally
3. People’s emotional expressions on Facebook predict friends’ emotional expressions, even days later

10

Facebook “News Feed”

11

Experiment Design

• Manipulated the extent to which people (N=689,003) were exposed to emotional expressions in their News Feed
• Test whether exposure to emotions led people to change their own posting behavior
• Participants: users who viewed Facebook in English

​

12

Experiment Design

• Posts were determined to be positive or negative using LIWC
• Each emotional post had between a 10% and 90% chance (based on their User ID) of being omitted from their News Feed for that specific viewing
• Treatment: X% of positive/negative posts omitted at random
• Control: (X%*percentage of positive/negative posts) of all posts omitted at random

13

Experiment Design

• One week experiment period (Jan 11-18, 2012)
• ~155,000 randomly selected participants based on User ID
• Dependent variable: the percentage of all words produced by a given person that was either positive or negative during the experimental period

14

Experiment Result

• Compare each emotion condition to its control
• Omitting emotional content reduced the amount of words the person subsequently produced
• Weighted linear regressions

15

Experiment Result

• Emotions expressed by friends, via online social networks, influence our own moods

16

Experiment Results

• Rules out alternative explanation of people interaction
• Contagion does not require nonverbal behavior
• Effect sizes when positivity and negativity were reduced is similar
• Withdrawal effect
• Effect sizes from the manipulations are small

17

Strengths

• Clever experiment design with clear identification strategies
• Large sample size with results showing differences between treatment and control group
• Rules out potential alternative explanations

18

Discussion & Critique

• Ethical concerns of the experiment design
• Definition of positive and negative posts could be improved
• The mechanisms behind emotional contagion remain unclear
• Could explore other emotions (sadness, anger, etc.)
• Quantification of the impact of emotional contagion and policy implications

19

2.

Rumor Cascades

Adrien Friggeri, Lada A. Adamic, Dean Eckles, Justin Cheng

20

General goal:

Examine the spread of rumors on Facebook comparing with Snopes.com

Main methods

“Uploading” and “Resharing”

Rumor Collecting

​

23

• Tracking spreading of rumor in facebook.

​

• Two types of information: a corpus of known rumors & sample of reshare cascades circulating on facebook.

​

• Matching the fb data to Snopes.com.

​

• Classified as True & False, also with range of values

Distribution of rumor in Snopes

24

Sampling Rumor Data

25

• Gathered 249,035 comments in photo from July to Aug 2013.

​

• Tagged 16,672 cascades with 62,487,651 shares.

​

• But the data is heavily biased.

​

• Public commenters tends to False than True in rumor judgements .

Biased Sample

26

• The fraction of comments linking to Snopes depends on the veracity of the rumor.

​

• Commenters are more likely to point out that a rumor is false than true

Presence and virality by category

​

27

• Both facebook and Snope possess their own properties in rumor spread and upload.

​

• Uploads on Facebook and Rumors in Snopes depends on certain categories.

​

• Fraction of Upload and Rumors also depends on certain categories.

​

• The average shares per upload and their categories are unrelated.

Uploads vs Shares

28

• Distribution of the number of shares of uploads before and after estimation

​

• Departure between estimated and observed.

Facebook vs Snope

29

• Certain categories were randomly assigned from snopes

​

• Higher distribution in facebook than snopes

​

Average share vs Category

30

• The virality and number of distinct cascades are unrelated.

​

• The average amount are really depends on the different Categories.

Cascade characteristics

​

31

• Two types of uploaded photo:users and pages

​

• The average amount are really depends on the different Categories.

​

• Consistent with the propagation replying less on who posts the rumor, but more on the highly contagious nature of the rumor itself

​

​

Reshare Deletion

32

• Effective in all three sorts of veracities, with different levels of significance.
• Not consistent within one kind of veracity, decided more by specific rumors.

Effects on resharing

33

• Snoping can motivate deletion if it comes in time.
• For false and true rumors, the resharing rates significantly decline after being snoped.
• The portion of reshares after and before snoping shows no great difference.
• *Snoping may slow the spread, but cannot terminate it.

Effects on resharing

34

• Snoping can motivate deletion if it comes in time.
• For false and true rumors, the resharing rates significantly decline after being snoped.
• The portion of reshares after and before snoping shows no great difference.
• *Snoping may slow the spread, but cannot terminate it.

45.2%

51.9%

59.4%

Rumor Variants and Burstiness

35

• Attracting variants of a successfully spread rumor will reburst in the future, reach new peaks with different time shifts.
• Rumor variants appear very early and stay dormant for some time before it flares up.

Rumor Variants and Burstiness

36

• Attracting variants of a successfully spread rumor will reburst in the future, reach new peaks with different time shifts.
• Rumor variants appear very early and stay dormant for some time before it flares up.

Rumor Variants and Burstiness

37

Counter-rumor

(465 members group)

• The bursting rumor variants often inherit the merits of its "parent".

Innocuous rumor recreation

(803 members group)

Counter-rumors

38

• Compared with messages directly dissuade people from doing something, parody triggers stronger popularity with more fun.
• A majority of spoof_1 (parody) reshares has seen the original rumor.

“[...] Facebook will start charging you for your account. To avoid this you must get naked [...]"

“[...] It even passed on TV. Facebook will start charging this summer. If you copy this on your wall [...]"

“don’t blindly copy and paste warnings just because your Facebook[...]"

Psychology of Rumor Spread

39

*Rumor Mills : The Social Impact of Rumor and Legend. / Fine, Gary A; Vincent, Veronique C; Heath, Chip.

Fact-finding Motivation

• People want to act effectively with knowledge of the environment.
• Rumors relate to processes of uncertainty reduction.

Psychology of Rumor Spread

40

*Rumor Mills : The Social Impact of Rumor and Legend. / Fine, Gary A; Vincent, Veronique C; Heath, Chip.

Relationship-Building Motivation

• The need of building and maintaining relationships motivates compliance and pleasing others.
• People transmit rumors to leave good impressions on others.

Psychology of Rumor Spread

41

*Rumor Mills : The Social Impact of Rumor and Legend. / Fine, Gary A; Vincent, Veronique C; Heath, Chip.

Self-Enhancement Motivation

• Self-enhancement refers to people's need in feeling positive about themselves, which therefore enhances self-defensiveness.
• Rumor "firms pre-existing attitudes rather than forming new ones".

Psychology of Rumor Spread

42

*Rumor Mills : The Social Impact of Rumor and Legend. / Fine, Gary A; Vincent, Veronique C; Heath, Chip.

Self-Enhancement Motivation

• Self-enhancement refers to people's need in feeling positive about themselves, which therefore enhances self-defensiveness.
• Rumor "firms pre-existing attitudes rather than forming new ones".

Strengths

• Using large volume of Facebook data and a reliable rumor database snopes.com to extract interesting characteristics of rumor cascades.
• A pioneering work in the study of rumor spread.

​

Weaknesses

• Collecting only the threads with reshares that contain links to snopes.com suffers from confounding bias.
• Only the variations in the volume of comments are analyzed, making it hard to identify the real effect of Snoping.
• This analysis doesn't consider other social media platforms.

43

Paper Discussions

Emotional Contagion

• How would you design the experiment with fewer ethical concerns?
• How can we quantify the effect of emotional contagion?

Rumor Cascades

• How to collect samples of rumors as comprehensive as possible?
• How could rumor flare-ups be ignited initially?
• What are the other reasons that motivate rumor spread?

44

Social Contagion Discussions

• What new opportunities can CSS bring in the research of social contagion?
• AI methods for text analysis, image & video, etc.
• Computational methods for social data (social network graphs, sharing & comments, etc.)
• How is social contagion different from before?
• New Channels (internet, social media) for social contagion to take place
• New groups and networks structures(Facebook groups, twitter topics, Reddit)

45

Thank you!

46