Information diffusion in online social networks

1
Information diffusion in online social networks
Lada Adamic School of Information, University of
Michigan, Ann Arbor
instant messaging
phones
email
blogs
online social networking sites
2
outline

• information diffusion in blogs
• strong ties
• viral marketing/recommendation networks
• expertise networks

3
Questions for this workshop

• How does network structure influence information
  diffusion?
• How does information diffusion shape networks?

4
formal organization (HPL labs)
5
Email correspondence information flow
6
Power laws and information spread on networks
94
6
2
Adamic, Lukose, Puniyani, Huberman, PRE 2001
7

• Information diffusion in blogs
• infer most likely information flow path

http//www-idl.hpl.hp.com/blogstuff
Adar, Zhang, Adamic, Lukose, WWE 2004
8
comment thread parody on Jim Henleys
Unqualified Offerings Blog
Communities and discourse
9
Can we understand community dynamics?

the political blogosphere, early 2005

• detecting polarization
• analyzing discourse
• learning what brings communities together online

Adamic Glance, LinkKDD 2005
10
1 Digbys Blog 2 James Walcott 3 Pandagon 4
blog.johnkerry.com 5 Oliver Willis 6 America
Blog 7 Crooked Timber 8 Daily Kos 9 American
Prospect 10 Eschaton 11 Wonkette 12 Talk Left 13
Political Wire 14 Talking Points Memo 15 Matthew
Yglesias 16 Washington Monthly 17 MyDD 18 Juan
Cole 19 Left Coaster 20 Bradford DeLong
21 JawaReport 22 Vodka Pundit 23 Roger L Simon 24
Tim Blair 25 Andrew Sullivan 26 Instapundit 27
Blogs for Bush 28 LittleGreenFootballs 29 Belmont
Club 30 Captains Quarters 31 Powerline 32 Hugh
Hewitt 33 INDC journal 34 Real Clear Politics 35
Winds of Change 36 Allahpundit 37 Michelle
Malkin 38 Wizbang 39 Deans World 40 Volokh
11
Liberals and conservatives differ in the topics
they discuss
Discussion of forged documents
12
How do memes evolve?
0200 AM Friday Mar. 05, 2004 PST Wired
publishes "Warning Blogs Can Be Infectious.
725 AM Friday Mar. 05, 2004 PST Slashdot posts
"Bloggers' Plagiarism Scientifically Proven"
955 AM Friday Mar. 05, 2004 PST Metafilter
announces "A good amount of bloggers are
outright thieves."
Before lunch Eytan writes FAQ Do bloggers kill
kittens?
After lunch Several bloggers title posts
Bloggers kill kittens!
13
Using strong ties to transmit or seek sensitive
info
Shi, Adamic, Strauss, Physica A, 2007
14
Structural definition of strong ties

• The strength of the tie is proportional to the
  number of closed triads it belongs to
• BuddyZoo dataset
• 140,000 AOL instant messenger users submitted
  buddy lists
• 8 million other users included in dataset

S3
15
strong ties permeate entire network?overlapping
communities
16
Networks and viral marketing diffusion with costs

• Some people recommend much more enthusiastically
  than others
• Recommendation cascades power-law distributed
• Hubs influence is limited
• Peer pressure is limited
• Links may weaken from overuse

Leskovec, Adamic, Huberman, EC 06, ACM Tweb 2007
product recommendation network
17
recommendation network for a single anime DVD

• purchase following a recommendation
• customer recommending a product
• customer not buying a recommended product

incentives can change the shape of a network
18
the data

• large anonymous online retailer (June 2001 to May
  2003)
• 15,646,121 recommendations
• 3,943,084 distinct customers
• 548,523 products recommended
• Products belonging to 4 product groups
• books
• DVDs
• music
• VHS

19
participation level by customer
very high variance
other power laws connected components, cascade
sizes, degree distribution
20
Network effects
21
How influential is peer pressure?
DVDs
BOOKS
22
does sending more recommendationsinfluence more
purchases?
DVDs
BOOKS
up to a point influence is limited to a couple
of dozen contacts
23
Use of social tie for recommendationsweakens the
tie
DVDs
BOOKS
24
recommendation success by book category

• lower than average success rate
• fiction romance, horror, etc. (1-2 success)
• not organized contexts spirituality (2-3
  success), tomato growing
• higher than average success rate
• professional technical medicine, engineering
  (4-6 success)
• other organized contexts Christianity (5),
  orchid growing

25
characteristics beyond category

26
Expertise networks in online communitiesstructu
re and algorithms
Jun Zhang, Mark Ackerman, Lada Adamic School of
Information, University of Michigan
27
motivation

• lots of people are turning to question-answer
  forums for help
• automatically infer the expertise of participants
• expertise could be used to rank answers, or
  recommend posts one could reply to

methods

• empirical evaluation of ranking algorithms
• social network analysis
• simulation
• understand underlying dynamics
• predict performance of ranking algorithms in
  communities with yet-unobserved dynamics

28
Constructing a community expertise network
unweighted
1
weighted by threads
2
Thread 1
Thread 2
1/2
weighted by shared credit
11//2
Thread 1 Large Data, binary search or hashtable?
user A Re Large... user B Re Large... user
C Thread 2 Binary file with ASCII data user
A Re File with... user C
weighted with backflow
0.1
29
JavaForum

• 87 sub-forums
• 1,438,053 messages
• community expertise network constructed
• 196,191 users
• 796,270 edges

Not Everyone Asks/Replies
The Web
The Java Forum
30
Uneven participation

• answer people may reply to thousands of others
• question people are also uneven in the number
  of repliers to their posts, but to a lesser
  extent

number of people one replied to
31
relating network structure to Java expertise

• Human-rated expertise levels
• 2 raters
• 135 JavaForum users with gt 10 posts
• inter-rater agreement (t 0.74, r 0.83)
• for evaluation of algorithms, omit users where
  raters disagreed by more than 1 level (t 0.80,
  r 0.83)

32
Structural Info Based Expertise Ranking Metrics

• replies posted ( answers)
• experts can answer many questions
• people replied to ( indegree)
• experts can answer questions from many different
  people
• z-score for the 2 above (observed m)/s
• experts are above the mean in the above two
  metrics
• PageRank replying to people who reply to people
• higher level experts can answer mid-level experts
• HITS experts answer questions by people whose
  questions other experts have answered

hubs point to good authorities
33
JavaForum empirical evaluation of ranking
algorithms
simple local measures do as well (and better)
than measures incorporating the wider network
topology
34
Modeling community structure to explain
algorithm performance
35
visualization
Best preferred
just better
36
degree correlation profiles
degree-degree correlations between asker and
helper
asker indegree
asker indegree
asker indegree
best preferred (simulation)
just better (simulation)
37
It can tell us when to use which algorithms
Preferred Helper best available
Preferred Helper just better
38
Summary

• Expertise Networks have interesting
  characteristics
• A set of useful metrics
• Simulation as an analysis tool
• There are rich design opportunities
• Find experts with the help of structural
  information (and content analysis)
• Predict good answers
• Re-order questions/answers to match expertise

questions posed by experts wait an average of 9
hours for the first reply novice questions are
answered in 40 minutes
working paper Expertise-Level based Interface
Personalization for Online Help-seeking
Communities
39
conclusions

• information diffuses on networks
• path is influenced by structure
• measured network is influenced by diffusion
• different interests/products bring people
  together
• incentives can modify social network structure
• positively (new connections)
• negatively (weakening existing connections)
• network is a reflection of the underlying
  expertise

40
for more info
Lada Adamic ladamic_at_umich.edu http//www-persona
l.umich.edu/ladamic