Influence and Correlation in Social Networks Xufei wang Nov-7-2008.
-
Upload
kimberly-merrin -
Category
Documents
-
view
214 -
download
0
Transcript of Influence and Correlation in Social Networks Xufei wang Nov-7-2008.
![Page 1: Influence and Correlation in Social Networks Xufei wang Nov-7-2008.](https://reader036.fdocuments.us/reader036/viewer/2022062806/56649c985503460f949543fe/html5/thumbnails/1.jpg)
Influence and Correlation in Social Networks
Xufei wangNov-7-2008
![Page 2: Influence and Correlation in Social Networks Xufei wang Nov-7-2008.](https://reader036.fdocuments.us/reader036/viewer/2022062806/56649c985503460f949543fe/html5/thumbnails/2.jpg)
2
Background, ConceptsProblem statementBasic ideaExperimental EvaluationFuture directions
Outline
![Page 3: Influence and Correlation in Social Networks Xufei wang Nov-7-2008.](https://reader036.fdocuments.us/reader036/viewer/2022062806/56649c985503460f949543fe/html5/thumbnails/3.jpg)
3
Proofs of social correlation
• People interact with others– Advices, reading, commenting– Communicating with others
• Non-causal correlation– Both the CO2 level and crime level have increased sharply– Both beer and diaper sales well in a super market
• Causal correlation– I bought an IPhone after I’m recommended by my friend
![Page 4: Influence and Correlation in Social Networks Xufei wang Nov-7-2008.](https://reader036.fdocuments.us/reader036/viewer/2022062806/56649c985503460f949543fe/html5/thumbnails/4.jpg)
4
Social influence
• A bought an IPhone after B told him it’s cool
– Directed: B influences A, not A influences B
– Chronological: A is influenced after B told him
– Asymmetry: B has influence to A doesn’t imply A has the same influence to B
![Page 5: Influence and Correlation in Social Networks Xufei wang Nov-7-2008.](https://reader036.fdocuments.us/reader036/viewer/2022062806/56649c985503460f949543fe/html5/thumbnails/5.jpg)
5
• Social influence: One person performing an action can cause her contacts to do the same.– A bought an IPhone after B told him it’s cool
• Homophily: Similar individuals are more likely to become friends.– Example: two mathematicians are more likely to
become friends.• Confounding factors: External influence from elements in
the environment.– Example: friends live in the same area, thus attend
and take pictures of similar events, and tag them with similar tags.
Sources of correlation
![Page 6: Influence and Correlation in Social Networks Xufei wang Nov-7-2008.](https://reader036.fdocuments.us/reader036/viewer/2022062806/56649c985503460f949543fe/html5/thumbnails/6.jpg)
6
Background, ConceptsProblem statementBasic ideaExperimental EvaluationFuture directions
Outline
![Page 7: Influence and Correlation in Social Networks Xufei wang Nov-7-2008.](https://reader036.fdocuments.us/reader036/viewer/2022062806/56649c985503460f949543fe/html5/thumbnails/7.jpg)
7
• Social correlation and social influence are different concepts
• Are they related?
• Maybe yes and Maybe no
Problem statement
![Page 8: Influence and Correlation in Social Networks Xufei wang Nov-7-2008.](https://reader036.fdocuments.us/reader036/viewer/2022062806/56649c985503460f949543fe/html5/thumbnails/8.jpg)
8
Background, ConceptsProblem statementBasic ideaExperimental EvaluationFuture directions
Outline
![Page 9: Influence and Correlation in Social Networks Xufei wang Nov-7-2008.](https://reader036.fdocuments.us/reader036/viewer/2022062806/56649c985503460f949543fe/html5/thumbnails/9.jpg)
9
• Influence model: each agent becomes active in each time step independently with probability p(a), where a is the # of active friends.
• Natural choice for p(a): logistic regression function:
with ln(a+1) as the explanatory variable. I.e.,
• Coefficient α measures social correlation.
Social correlation evaluation
![Page 10: Influence and Correlation in Social Networks Xufei wang Nov-7-2008.](https://reader036.fdocuments.us/reader036/viewer/2022062806/56649c985503460f949543fe/html5/thumbnails/10.jpg)
10
• Shuffle Test: – Chronological property
• Edge-Reversal Test: – Asymmetry property
Testing for influence
User A B C
Time 1 2 3
User A B C
Time 2 3 1
A
B
C
A
B
C
![Page 11: Influence and Correlation in Social Networks Xufei wang Nov-7-2008.](https://reader036.fdocuments.us/reader036/viewer/2022062806/56649c985503460f949543fe/html5/thumbnails/11.jpg)
11
Background, ConceptsProblem statementBasic ideaExperimental EvaluationFuture directions
Outline
![Page 12: Influence and Correlation in Social Networks Xufei wang Nov-7-2008.](https://reader036.fdocuments.us/reader036/viewer/2022062806/56649c985503460f949543fe/html5/thumbnails/12.jpg)
12
• Influence model– Only use the influence factor– Current node A has “a” active friends, its probability to
be active is related with the # of active friends
• Correlation model– Use the homophily and confounding factors – Init S nodes as centers randomly, add a ball of radius 2
to each node in S, according to the data on Flickr, randomly pick the same # of nodes to be active
Experimental setup
![Page 13: Influence and Correlation in Social Networks Xufei wang Nov-7-2008.](https://reader036.fdocuments.us/reader036/viewer/2022062806/56649c985503460f949543fe/html5/thumbnails/13.jpg)
13
Simulation results Shuffle test, influence model
![Page 14: Influence and Correlation in Social Networks Xufei wang Nov-7-2008.](https://reader036.fdocuments.us/reader036/viewer/2022062806/56649c985503460f949543fe/html5/thumbnails/14.jpg)
14
Simulation results Edge-reversal test, influence model
![Page 15: Influence and Correlation in Social Networks Xufei wang Nov-7-2008.](https://reader036.fdocuments.us/reader036/viewer/2022062806/56649c985503460f949543fe/html5/thumbnails/15.jpg)
15
Simulation results Shuffle test, correlation model
![Page 16: Influence and Correlation in Social Networks Xufei wang Nov-7-2008.](https://reader036.fdocuments.us/reader036/viewer/2022062806/56649c985503460f949543fe/html5/thumbnails/16.jpg)
16
Simulation results Edge-reversal test, correlation model
![Page 17: Influence and Correlation in Social Networks Xufei wang Nov-7-2008.](https://reader036.fdocuments.us/reader036/viewer/2022062806/56649c985503460f949543fe/html5/thumbnails/17.jpg)
17
Shuffle test on Flickr data
![Page 18: Influence and Correlation in Social Networks Xufei wang Nov-7-2008.](https://reader036.fdocuments.us/reader036/viewer/2022062806/56649c985503460f949543fe/html5/thumbnails/18.jpg)
18
Edge-reversal test on Flickr data
![Page 19: Influence and Correlation in Social Networks Xufei wang Nov-7-2008.](https://reader036.fdocuments.us/reader036/viewer/2022062806/56649c985503460f949543fe/html5/thumbnails/19.jpg)
19
Explanations
• The users’ tagging actions are independent
• The users either seldom visit their friends’ pages
• Or the users visit pages but only care about the content rather than the tags
![Page 20: Influence and Correlation in Social Networks Xufei wang Nov-7-2008.](https://reader036.fdocuments.us/reader036/viewer/2022062806/56649c985503460f949543fe/html5/thumbnails/20.jpg)
20
Background, ConceptsProblem statementBasic ideaExperimental EvaluationFuture directions
Outline
![Page 21: Influence and Correlation in Social Networks Xufei wang Nov-7-2008.](https://reader036.fdocuments.us/reader036/viewer/2022062806/56649c985503460f949543fe/html5/thumbnails/21.jpg)
21
Future directions I
• The relationship in the internet is weak!– How weak it is?
• So I think it’s interesting to search close communities, based on strong correlation, in blogosphere– How to define the “strongness”– How the “strongness” among the users– Do we have reasonable datasets– “strongness” is related with time?
![Page 22: Influence and Correlation in Social Networks Xufei wang Nov-7-2008.](https://reader036.fdocuments.us/reader036/viewer/2022062806/56649c985503460f949543fe/html5/thumbnails/22.jpg)
22
Future Directions II
• Most of the users don’t contact frequently– How about the contact distribution
• Search for stable relationships is also interesting. Seeking stable communities– How to define stable?– Stable relationship can be strong or weak connection– Contact infrequently but regularly– The group can be small– Hold for a long time??