Influence on Twitter

1
Influence on Twitter

• Presented by Ruhsary Rexit
• Nov 22, 2011
• CS Department
• University of Pittsburgh

2
Reference

• Everyones an Influencer Quantifying Influence
  on Twitter
• B. Eytan, H. Jake, M. Winter, and W. Duncan.
  WSDM11, Hong Kong, China, 2011
• What is Twitter, a Social Network or a News
  Media?
• H. Kwak, C. Lee, H. Park, and S. Moon. Pages
  591600. ACM, 2010.
• Measuring User Influence on Twitter The Million
  Followers Fallacy
• M. Cha, H. Haddadi, F. Benevenuto, and K. P.
  Gummad. In 4th Intl AAAI Conference
• on Weblogs and Social Media, Washington, DC, 2010.

3
Outline

• Problem definition
• Motivation
• Techniques and results
• Question

4
The definition of Influence

• Dictionary meaning of Influence
• The capacity or power of persons or things to
  produce effects on the actions, behavior,
  opinions of someone else.

5
Influence of Twitter

• If we can define "A influenced B" then
• Influence of a seed post if there is a given
  seed node with a URL, how many nodes are
  influenced in the follower graph
• Influence of an individual average influence of
  seed post by him/her

6
The definition of A influenced B

• If B follows A
• A posted the url earlier than B
• A is the only Bs friend to post the url
• If B has multiple friends who posted the URL
• Influencers to be the first, or the latest, split
  equally
• Almost the same for the above three in
  experiments

7
Three ways of assigning influence to multiple
sources
8
Information diffusion

• Word-of-Mouth diffusion
• an important mechanism influencing large
  population
• Diffusion on twitter?
• Copying post (Repost) on Twitter
• More specifically, Repost the same URL

9
Example of Information Diffusion
10
Object

• Predict one's influence
• How to use influence for marketing

11
Motivation

• Why twitter?
• The network over word-of-mouth is usually
  difficult to observe/record, influence is
  difficult to attribute accurately
• Observational data on diffusion are heavily
  biased toward "successful" diffusion events,
  i.e., only the successful diffusions will be
  recorded

12

• Twitter/micro-blogs provides a natural laboratory
  for study of diffusion process, because Twitter
  is devoted to disseminating information.
• Users frequently like to share web-content, since
  twitter post short, restricted to 140 characters,
  easy to collect and study large scale network.

13
Influence

• On real life
• Trusted friends definitely have influence
• Type of an individual who considered special
• experts, journalists, public figures, media
  representativeness

14
Influence

• On Twitter
• no need for labeling
• Influence of an individual is possible to measure
  and compare by their activity which is observable
  on Twitter itself.
• no ambiguity
• twitter has a useful future, which is all the
  users communicate almost same way by tweeting to
  their followers.
• not difficult to compute
• Influence on twitter can be quantified by using a
  given post by number of users who repost the URL,
  and they can be traced back to the first user
  posted this through follower graph.

15
Related Work---Ref 2

• Examine influence and diffusion on Twitter
• Compared three different measures of Influence
• number of followers
• page-ranking
• number of re-tweets
• Concluded that the ranking of the most
  influential users differed depending on measure

16
Related Work---Ref3

• Compared three different measures
• number of followers
• number of re-tweets
• number of mentions
• Concluded that the most followed users did not
  necessarily score highest.

17
Current work

• Measure influence in terms of the size of entire
  diffusion tree associated with each event (URL)
• Instead of most followed users, consider all
  users
• In addition to predicting diffusion of the
  attributes of individual seeds, also study the
  effects of content

18
Experiment Data

• Time period two month 9/13 -11/15 2009
• Influenced users 1.6 M
• total tweets 1.3 B
• total URL (bit.ly) 86 M
• Diffusion event 74 M (initiated by seed users
  who were active in two months)

19
(No Transcript)
20
How to predict the influences

• Seed user attributes
• followers, friends, tweets, date of joining
• Past influence of a seed user
• average minimum, maximum total influence (the
  average number of reposts by that user's
  immediate followers in the first month of the
  observation period)
• average minimum, maximum local influence (the
  average total cascade size over the same period)

21
Result
22
Result
23
Applying influence on marketing

• Cost Function

24
Question ???

• Applying influence on other events except from
  marketing
• Spreading important event, news, urgent needs
• Other social networks except from twitter, such
  as
• facebook, youtube, flickr, etc.