Dynamics of Communication in Online Social Networks · 2013. 8. 15. · Ajita John 2and...
52
June 21, 2009 1 Dynamics of Communication in Online Social Networks Munmun De Choudhury 1 , Hari Sundaram 1 , Ajita John 2 and Dorée Duncan Seligmann 2 1 Arts, Media and Engineering, Arizona State University 2 Avaya Labs Research, NJ @Avaya Labs
Transcript of Dynamics of Communication in Online Social Networks · 2013. 8. 15. · Ajita John 2and...
June 21, 2009 1
Dynamics of Communication in Online
Social Networks
Munmun De Choudhury1, Hari Sundaram1,
Ajita John2 and Dorée Duncan Seligmann2
1Arts, Media and Engineering, Arizona State University2Avaya Labs Research, NJ
@Avaya Labs
June 21, 2009 2
Outline
‘Interestingness’ of
conversations
‘Social Synchrony’ of
user actions
Why do we care about
social communication?
Conclusions &
Future Directions
June 21, 2009@Avaya Labs 2
Introduction
June 21, 2009 3
I want to buy a new digital SLR…
help!
June 21, 2009@Avaya Labs 3
June 21, 2009 4
I want to have a high school get-
together… should I call Jake or
Mary to try and contact the
maximum number of people?
June 21, 2009@Avaya Labs 4
June 21, 2009 5
What is likely to be the stock price
next week?
June 21, 2009@Avaya Labs 5
June 21, 2009 6
Why do we care about
communication in social networks?
June 21, 2009@Avaya Labs 6
June 21, 2009 7
Current State-of-the-Art…
@Avaya Labs
Face-to-faceIP Telephony
EmailsInstant Messaging
Text Messaging
June 21, 2009 8
The New Modes of Communication…
Slashdot
Engadget
Flickr
LiveJournalDigg
YouTubeBlogger
MetaFilterReddit
MySpaceOrkut
June 21, 2009 9
Some Statistics…
19,128,882 accounts.Live Journal
8.9M visitors.Huffington Post
1,887,887 monthly visitors.Engadget
3M unique users; $40M.Digg
110M monthly active users; 14B comments on
the site.
MySpace
200M active users; 1B pieces of content (web
links, news stories, blog posts, notes, photos,
etc) shared each week.
3.6B images; 50M users.Flickr
139M users; US$200M [Forbes].YouTube
@Avaya Labs
June 21, 2009 10
So what do these imply?
� From a sociological perspective,
• Diffusion – “word of mouth”
• The process of the passing of information from person to
person.
• Viral marketing, buzz based recommendations
• Network evolution
• Community emergence
• Correlation of social processes with external
phenomena
• Reflection on individual social behavior
• Emergence of new social semantics
June 21, 2009 11June 21, 2009@Avaya Labs 11
A news reporter A political analyst A company You…
Who could benefit from this research?
June 21, 2009 12
Research Questions
� Communication and emergent local properties
• How does communication impact flow of information?
• How does synchrony emerge in social media due to
communication?
� Communication and emergent global properties
• How do groups emerge and evolve as an artifact of
communication?
� Communication and emergent media properties
• What are the emergent media properties that result from
communication on shared media?
@Avaya Labs
June 21, 2009 13
Our Work: Overview
Analysis of DynamicsAnalysis of Patterns
WWW 2009 (11%)
Submission to ACM TOMCCAP
ICME 2009 (22%)Rich Media
SocialCom 2009 (9%)Future WorkNetworks
HT 2008 (24%)CIKM 2008 (16%)
HT 2009
Submission to ACM TOIS
Groups
WI 2007 (17%)
HT 2008 (24%)
Submission to ACM T-Web
GHC 2008Individuals
@Avaya Labs
June 21, 2009 14
Characterizing Communication
through ‘Interesting’ Conversations
June 21, 2009@Avaya Labs 14
June 21, 2009 15
Why do people repeatedly come
back to the same You Tube video?
June 21, 2009@Avaya Labs 15
June 21, 2009 16
They are returning to do more than
watch the same video
June 21, 2009@Avaya Labs 16
June 21, 2009 17
We think it is the conversations
around the video they find
interestingJune 21, 2009@Avaya Labs 17
June 21, 2009 18
What is a Conversation?
� A conversation associated with an online social media object is a
temporally ordered sequence of comments posted by individuals.
June 21, 2009@Avaya Labs 18
June 21, 2009 19
Themes and participants make
conversations interesting
June 21, 2009@Avaya Labs 19
June 21, 2009 20
An example of an interesting conversation …
June 21, 2009@Avaya Labs 20June 21, 2009 20
June 21, 2009 21
Conversational interestingness is not
necessarily the popularity of a media
object or preference of a topic
June 21, 2009@Avaya Labs 21
June 21, 2009 22June 21, 2009@Avaya Labs 22
� Goal:• What causes a conversation to be interesting, that prompts a
user to participate in the discussion on a posted video?
Our Contributions
June 21, 2009 22
� Approach:
• Detect conversational themes.
• Determine interestingness of
participants and interestingness of
conversations based on a random
walk model.
• Measure the consequence of a
conversations.
• Excellent results on a dataset
from YouTube.
June 21, 2009 23June 21, 2009@Avaya Labs 23
Conversational Themes
� Conversational themes are sets of salient topics associated with
conversations at different points in time.
bag of words (chunks, λi,t) over time slices
t1 t2 tQ
……
( )θ λ ,| ,j i t
p t
……Theme models, θj
conversation i = ?
June 21, 2009 24
Interestingness of Conversations
ψ 1−ψImpact due to
participantsImpact due to
themes
Interestingness of
participants
Relationship between
participants and
conversations
Conversational
Theme Distribution
Theme Strength
Interestingness of
conversations
X
X
X
Interestingness of
conversations
t─1 t
ψ
1─ψ
June 21, 2009@Avaya Labs 24
June 21, 2009 25June 21, 2009@Avaya Labs 25
Interestingness of Participants
β1−β
Past communication Preference
June 21, 2009 26
Interestingness of participants and
conversations mutually reinforce
each other
June 21, 2009@Avaya Labs 26
June 21, 2009 27June 21, 2009@Avaya Labs 27
Joint Optimization of Interestingness
� A joint optimization framework, which maximizes the two
interestingness measures for optimal X=(α1, α2, α3, ψ) and also
incorporates temporal smoothness:
( ) ( ) ( ) ( )ρ ρ= + − + − + −2 2
1( ) . (1 ). exp expP C
g d dP C
X I X I X
Interestingness of
participantsInterestingness of
conversations
Regularization of
participants’
interestingness
Regularization of
conversations’
interestingness
June 21, 2009 28
Three consequence metrics of
interestingnessJune 21, 2009@Avaya Labs 28
t t+δ
act
ivit
y
Participant
activityt t+δ
participant cohesiveness
Participant
Cohesiveness
t t+δ
conversation
Thematic
interestingness
June 21, 2009 29June 21, 2009@Avaya Labs 29
June 21, 2009 30June 21, 2009@Avaya Labs 30
Analysis of Interestingness of Conversations
� Mean interestingness of conversations increases during periods of several
external events; however, certain highly interesting conversations always occur
at different weeks irrespective of events.
June 21, 2009 31
Evaluation using Consequences
� Interestingness is computed using five techniques –• our method with temporal smoothing (I1),
• our method without temporal smoothing (I2) and
• the three baseline methods, • B1 (comment frequency),
• B2 (novelty of participation),
• B3 (co-participation based PageRank).
June 21, 2009@Avaya Labs 31June 21, 2009 31
June 21, 2009 32
Summary…
June 21, 2009@Avaya Labs 32
June 21, 2009 33
Characterizing Communication
through ‘Synchrony’ of User Actions
June 21, 2009@Avaya Labs 33
June 21, 2009 34
Clapping in an Auditorium
@Avaya Labs
June 21, 2009 35
Biological Oscillators
@Avaya Labs
June 21, 2009 36
Movement of herds of animals
@Avaya Labs
June 21, 2009 37
What causes users on a social media
mimic each other with respect to a
certain action?
@Avaya Labs
June 21, 2009 38
Observations from Digg
Topic ‘Olympics’ is observed to exhibit synchrony where old users continue to be involved in
the action of digging stories, as well as large number of new users join in the course of time
(Sept 3-Sept 13).
@Avaya Labs
June 21, 2009 39
Defining Social Synchrony…
� Social synchrony is a temporal phenomenon occurring in social networks which is characterized by:• a certain topic
• an agreed upon action
• a set of seed users involved in performing the action at a certain point in time, and
• large numbers of continuing old users as well as new users getting involved over a period of time in the future, following the actions of the seed set.
@Avaya Labs
June 21, 2009 40
The distinction with information
cascades…
June 21, 2009@Avaya Labs 40
Leskovec et al 2007
June 21, 2009 41
Our Contributions
� Goal:
• a framework for predicting social synchrony in
online social media over a period of time into the
future.
� Approach:
• Operational definition of social synchrony.
• Learning – a dynamic Bayesian representation of
user actions based on latent states and
contextual variables.
• Evolution – evolve the social network size and
the user models over a set of future time slices
to predict social synchrony.
� Excellent results on a large dataset from the
popular news-sharing social media Digg.
@Avaya Labs
June 21, 2009 42
Main Idea
� Socially-aware and unaware states.
� Learning – for each user in the social network, we need to predict her probability of actions at each future time slice.
� Evolution –synchrony in a social network (a) is likely to involve sustained participation; and (b) persists over a period of time.• Evolve network
• Evolve user models
• Predict synchrony
@Avaya Labs
June 21, 2009 43
The Learning Framework
� A user’s intent to perform an action depends upon her state.
� The user state in turn is affected by the user context (e.g. actions of
the neighboring contacts, coupling with seed users and / or the
user’s communication over the topic).
@Avaya Labs
June 21, 2009 44
Estimation
( ) ( ) ( )
( ) ( )
− − − − − −
− −
= ⋅
= ⋅
∑
∑
� � �� � �
,
,
, , 1 , 1 , , , 1 , 1 , , 1 , 1
, , , , 1 , 1
| , | , , | ,
| | , ,
u j
u j
u j u j u j u j u j u j u j u j u j u j
S
u j u j u j u j u j
S
P A A P A S A P S A
P A S P S S
C C C
C
Estimate user context
Estimate probability of
user state given
context
Estimate probability of
user action given the
state
@Avaya Labs
A continuous Hidden Markov
Model where the actions are
the emissions
Multinomial density of states
over the contextual attributes
with a Dirichlet prior
where,
Au,j= action of user u at time slice j
Cu,j-1= context of user u at time slice j-1
Su,j= state of user u at time slice j
June 21, 2009 45
The Evolution Framework
� Why?
• Online learning methods (e.g. incremental SVM Regression) that
incrementally train and predict a value at each time slice, are
not helpful.
• Synchrony needs to be predicted over a set of future time slices.
� Method:
• Estimating network size
• Evolving user models
• Choosing users based on high probability of comments / replies
• Predicting synchrony
@Avaya Labs
June 21, 2009 46
Experiments on Prediction
@Avaya Labs
� Digg dataset
• August, September 2008
• 21,919 users, 187,277 stories,
7,622,678 diggs, 687,616
comments and 477,320 replies.
• Six sample topics – four
inherently observed to have
synchrony.
June 21, 2009 47
Comparative Empirical Study
� Baseline methods:
• B1: temporal trend learning method of user actions
• B2: a linear regressor based method over users’ comments and replies
• B3: SIR (susceptible-infected-removed) epidemiological model
• B4: a threshold based model of global cascades
@Avaya Labs
0.270.320.410.530.15Tennis
0.220.290.360.490.12Celebrity
0.270.310.40.460.13Comedy
0.410.440.490.540.19Olympics
0.280.290.360.410.11World News
0.350.380.520.670.19US Elections
B4
B3
B2
B1
Our MethodTopics
June 21, 2009 48
Summary…
June 21, 2009@Avaya Labs 48
June 21, 2009 49
Conclusions and Future Directions
June 21, 2009@Avaya Labs 49
June 21, 2009 50
Conclusions
� Communication is central to the
evolution of social systems.
� Impact on sociological, business and
theoretical domains.
� Modeling ‘interestingness’ of
conversations:
• Themes, participants and consequences
� Modeling social ‘synchrony’:
• User actions, latent states of social
awareness and unawareness, and user
context
@Avaya Labs
June 21, 2009 51
Future Directions
Network evolution – birth and death of communities
Comprehensive model of human communication behavior
Pro
toty
pe
Ap
pli
cati
on
s!!
@Avaya Labs
