Friendship Paradox Redux, ICWSM 2013

• Confirm and Explore the Friendship Paradox on Twitter
• Demonstrate New Paradoxes: Activity Paradox and Virality Paradox
• Implications: Information Overload and Altered Propagation

Does the Friendship Paradox hold on Twitter in all directions?

476 Million Tweets (Random Sample of Firehose) from June to December 2009 (Yang and Leskovec 2011)

+

Follower Graph of 40 Million Users as of Summer 2009 (Kwak et al. 2010)

3.4 Million Users sending 37 Million Tweets

Paradox worsens with connectivity

Overloaded users SEND and RECEIVE more popular content

Your Friends and Followers...

• have more Friends and Followers,
• are more Active,
• and send and receive more Viral content than you.

"Your friends are more interesting than you are!"

Incoming information scales super-linearly with the number of friends

Overloaded Users receive a more popular content than Underloaded Users, on average

Implications

• Users are not uniformly sampling interesting content
• Unexplained systematic biases stem from connectivity-activity correlations

• Increasing engagement with social media exacerbates observed paradoxes.
• Decreasing engagement decreases exposure to popular content
• Challenge: build compelling social network w/o information overload

Our Contribution

Friendship Paradox

The Friendship Paradox⟩

Feld, American Journal of Sociology 1991

Related Work

friendship paradox

*One direction only

• In "meatspace" (Feld 1991; Zuckerman and Jost 2001)
• Facebook (Ugander et al. 2011)
• Twitter* (Garcia-Herranz et al. 2012)

friends as detectors

• Virtual Outbreaks (Garcia-Herranz et al. 2012)
• Pathogenic Outbreaks (Christakis and Fowler 2010)

social perception and decision making

• Alcohol and Drug Abuse (Tucker et al. 2011; Wolfson 2000)
• Wealth (Morselli and Tremblay 2004; Amuendo-Dorantes 2007)
• Extraversion (Pollet et al. 2011; Querca et al. 2012)

The Friendship Paradox holds in Every Direction!

"friends are better connected"

99.6%

"followers are better connected"

98%

99%

shaded region corresponds to paradox condition

"friends are more popular"

98%

"followers are more popular"

New Paradoxes

Is information overlaod inevitable?

Virality Paradox

Flux of Incoming Information Grows Faster Than User Engagement

76%

Ratio of Cascade Sizes Received

Activity Paradox

86%

Ratio of Cascade Sizes Sent

88% or 99%

"Your friends send and receive higher virality content than you do."

Correlation between Number of Friends and Activity

Avg. Posted Tweets Per Friend / Posted Tweets by User

Friendship Paradoxes on Twitter

"Your friends are more active than you are."

Friendship Paradox Redux:

Your Friends are More Interesting Than You

Consequence of Information Overload

Incoming Tweets Scale Super-Linearly!

Finding "Overloaded Users"

Bin users into activity levels:

<5 Tweets over 2-month window

5-19 Tweets

20-59 Tweets

>= 60 Tweets

• Top 33% of users in each category, based on received tweets, we call "Overloaded"
• Bottom 33% of users in each category, based on received tweets, we call "Underloaded"

Mean of average size of received URL cascades

Dilemma: Popularity vs. Overload

Least Active Users

Most Active Users

Paradoxes and

Information Overload

Conclusions

Friendship Paradoxes on Twitter:

Does Directionality Matter?

Friend = Followee

5.8 Million Users w/ ~200 Million links

Users active over a two month window

Nathan Hodas

Farshad Kooti

Kristina Lerman

Information Sciences Insitute

University of Southern California

Many Thanks

Co-Authors

Researchers

• Farshad Kooti
• Prof. Kristina Lerman

Funders

Practitioners

This material is based upon work supported in part by the Air Force Office of Scientific Research under Contract Nos. FA9550-10-1-0569, by the National Science Foundation under Grant No. CIF-1217605, and by DARPA under Contract No. W911NF-12-1-0034.