W HO G IVES A T WEET ? Evaluating Microblog Content Value Paul André @paulesque Michael Bernstein Kurt Luther Carnegie Mellon & Uni. Southampton MIT CSAIL.

WHO GIVES A TWEET?Evaluating Microblog Content Value

Paul André@paulesque

Michael

BernsteinKurt Luther

Carnegie Mellon & Uni. Southampton

MIT CSAILGeorgia Institute of Technology

?

?

What content is valued, and why?

?

What content is valued, and why?

1. design implications

2. emerging norms and practice

DESIGN

Who Gives a Tweet?anonymous feedback from followers and strangers

(analysis of follower ratings only)

DESIGN

anticipated reciprocity

Who Gives a Tweet?anonymous feedback from followers and strangers

rate tweets(provide us data)

receive value in return(ratings from followers)

DESIGN

wgat_user:

username:

RECRUITMENT

RECRUITMENT

RECRUITMENT

1,443 users

rated 43,738 tweets

from 21,014 Twitter

accounts

entire dataset

RESULTS

36% Worth Reading39% Neutral

25% Not Worth

Reading

41% Worth Reading

average

user

What content is valued,and why?

What content is valued,and why?

1. categories

2. reasons why

What content is valued,and why?

4,220 tweets

Ground truth +

CrowdFlowerCohen’s Kappa: 0.62

Category labelsmore Information Sharing (49% vs 22%)

less Me Now (10% vs 40%)

+ inclusion of organizations

compared to random sample in Naaman

(2010)

RESULTS: CategoriesPredictor

Question to Followers

Information Sharing

Self-Promotion

Random Thought

Opinion / Complaint

Me Now

Conversation

Presence Maintenance

RESULTS: CategoriesPredictor

Question to Followers

Information Sharing

Self-Promotion

Random Thought

Opinion / Complaint

Me Now

Conversation

Presence Maintenance

“gud morning twits”

20%liked

45%disliked

RESULTS: CategoriesPredictor

Question to Followers

Information Sharing

Self-Promotion

Random Thought

Opinion / Complaint

Me Now

Conversation

Presence Maintenance

Odds Ratio

2.83

2.69

2.69

2.47

2.05

1.89

1.57

N/A

“gud morning twits”

20%liked

45%disliked

*p<.01˘trend p=.05

Odds Ratio

2.83

2.69

2.69

2.47

2.05

1.89

1.57

N/A

RESULTS: CategoriesPredictor

Question to Followers

Information Sharing

Self-Promotion

Random Thought

Opinion / Complaint

Me Now

Conversation

Presence Maintenance

“What'd they say?? @adam807 Dreamed I went to an @waitwait taping and they had to stop because a guest made @petersagal cry.”24%liked

34%disliked

*p<.01˘trend p=.05

Odds Ratio

2.83

2.69

2.69

2.47

2.05

1.89˘

1.57

N/A

RESULTS: CategoriesPredictor

Question to Followers

Information Sharing

Self-Promotion

Random Thought

Opinion / Complaint

Me Now

Conversation

Presence Maintenance

“tired and upset”

27%liked

25%disliked

*p<.01˘trend p=.05

Odds Ratio

2.83*

2.69*

2.69*

2.47*

2.05˘

1.89˘

1.57

N/A

RESULTS: CategoriesPredictor

Question to Followers

Information Sharing

Self-Promotion

Random Thought

Opinion / Complaint

Me Now

Conversation

Presence Maintenance

*p<.01˘trend p=.05

Odds Ratio

2.83*

2.69*

2.69*

2.47*

2.05˘

1.89˘

1.57

N/A

RESULTS: CategoriesPredictor

Question to Followers

Information Sharing

Self-Promotion

Random Thought

Opinion / Complaint

Me Now

Conversation

Presence Maintenance

*p<.01˘trend p=.05

Odds Ratio

2.83*

2.69*

2.69*

2.47*

2.05˘

1.89˘

1.57

N/A

RESULTS: CategoriesPredictor

Question to Followers

Information Sharing

Self-Promotion

Random Thought

Opinion / Complaint

Me Now

Conversation

Presence Maintenance

*p<.01˘trend p=.05

Not Worth Reading

RESULTS: Reasons

Not Worth Reading

Old News “Yes, I saw that first thing this

morning.”

“Since your followers read the

NYT too, reposting NYT URLs

is tricky unless you add

something.”

No Personal Touch

Conversations “Twitter’s fault; feels like

listening in on a private

conversation”

RESULTS: Reasons

Not Worth Reading

Banal or ProsaicTweets

“…and so what?”

“Just links are the worst thing in

the world.”

Lack of Context

Professional vs Personal Insight

“I unfollowed you for this tweet. I

don’t know you; I followed you b/c

of you job.”No Curiosity “All the news I need is here. Not

much of a tease.”

RESULTS: Reasons

Worth Reading

RESULTS: Reasons

Worth Reading

Valued Information

“interesting perspective on

something I know nothing about.”

“makes you want to know more.”Appealing Description

Conciseness “few words to say much, very

clear.”Human “personal, honest, and

transparent.”

RESULTS: Reasons

Embed more context in tweets (be less

cryptic)

Add extra commentary, especially if

RTing

Use twitter-specific mechanisms

(hashtags, @mentions, and DMs)

appropriately

Unique hashtag for questions is

valued

Conciseness, even with 140 chars,

valued

Happy sentiments valued; whining

disliked

IMPLICATIONS FOR PRACTICE

Exploring different communities on Twitter

Which results generalize

Rate author, not tweet

Users no longer followed

Self-ratings

Twitter as maintaining awareness and

relationships

LIMITATIONS

FUTURE WORK

DISCUSSION

Utilizing

results:

Twitter’s simplicity vs. Facebook’s newsfeed complexity

Presentation:

Technological intervention:

design tools to learn, filter, re-present

Social intervention:

inform users of perceived value and

reaction

Social media

sites: but also new questions of

content value and accepted practice

new connection opportunities

Design sites to elicit more subtle reactions

Sample of 1,400 users and 43,000 ratings:

CONCLUSIONS

41% of feed worth reading

Information Sharing liked / Me Now

disliked

Reasons: context, commentary,

conciseness, …

Technological and social interventions

Social media

sites: but also new questions of

content value and accepted practice

new connection opportunities

Design sites to elicit more subtle reactions

Sample of 1,400 users and 43,000 ratings:

41% of feed worth reading

Information Sharing liked / Me Now

disliked

Reasons: context, commentary,

conciseness, …

Technological and social interventionsCONCLUSIONSCONCLUSIONSCONCLUSIONSThanks for listening!with thanks to Ed Cutrell, Robert Kraut, m.c. schraefel, Ryen White, Sarita Yardi, HCII Social Comp. group and anonymous reviewers

Paul André – CMU HCIIMichael Bernstein – MIT CSAILKurt Luther – Georgia Tech GVU

RESULTSCategoriesPredictor Odds

Ratioz value

Question to Followers 2.83 2.94*

Information Sharing 2.69 3.05*

Self-Promotion 2.69 2.61*

Random Thought 2.47 2.89*

Opinion / Complaint 2.05 1.93˘

Me Now 1.89 1.94˘

Conversation 1.57 1.26

Presence Maintenance N/A N/A

RESULTSCategoriesQuestion to Followers

Information Sharing

Self-Promotion

Random Thought

Opinion / Complaint

Me Now

Conversation

Presence Maintenance

47% chance of being Worth Reading

“This is a good use of Twitter.”

“Gives one pause to think about the question posted.”

Questions to Followers

RESULTSCategoriesQuestion to Followers

Information Sharing

Self-Promotion

Random Thought

Opinion / Complaint

Me Now

Conversation

Presence Maintenance

“The headline arouses my curiosity.”

“Wow. Didn’t know that was happening. Thanks for informing me.”

Information Sharing

RESULTSCategoriesQuestion to Followers

Information Sharing

Self-Promotion

Random Thought

Opinion / Complaint

Me Now

Conversation

Presence Maintenance

22% chance of being Worth Reading

“Sorry, but I don’t care what people are eating.”

“Too much personal info.”

“He moans about this ALL THE TIME. Seriously.”

Me Now

RESULTSCategoriesQuestion to Followers

Information Sharing

Self-Promotion

Random Thought

Opinion / Complaint

Me Now

Conversation

Presence Maintenance

Me Now “Foursquare updates don’t need to be

shared on Twitter unless there’s a

relevant update to be made.”

“4sq, ffs.”

RECRUITMENT