Managing Social Communities

Web Science & Technologies
University of Koblenz ▪ Landau, Germany

Managing Social Communities

Steffen Staab

Acknowledgements to ROBUST Project team & WEST Team,
in particular
K. Dellschaft, J. Kunegis, F. Schwagereit

Institut WeST – Web Science & Technologies

Semantic Web Web Retrieval Interactive Web Multimedia Web Software Web

eGovernment eMedia eScience eOrganizations ePerson

Institute for Computer Institute for Leibniz Institute for
Science Information Systems Social Sciences (GESIS)
Steffen Staab Web Science Doctoral
staab@uni-koblenz.de Summer School 2

Plan for this Talk

1 Web

2 Science


Social Communities

…are everywhere

c


Risks Opportunities
Bad content quality, Open innovation,
social ill behavior,… improved user support,…
 jeopardize business value  increase business value

Data Storage Content, User &
and Processing Networks Analysis
Scalability, heterogeneity Understanding,
response time

Business Value
Product support & innovation, CRM, Expertise management, Marketing, Advertising
Online Communities Intranet, Extranet, Internet

Large-scale Testbeds

2013 2013
5M users millions posts/day
1200K accesses/day 1TB data/day
SAP (B2B) Polecat (C2C)
Community Network
Online Marketing
Business Partner Network
CRM for IT
2009
2009
1.5M users
…
150K access/day

IBM (E2E)
Developer Network
2009 2013
Corporate Knowledge
99K accounts 800K accounts
Management

2 staab@uni-koblenz.de Summer School 6

SAP Business Partner Use Case

SAP Developer Network

Size of user generated
Posts per day Number of users
content (posts)
2007 2009 2013 2007 2009 2013 2007 2009 2013

SAP 5000 6000 7000 1M 4M 10.0 1M 1.7M 4.8M
M


ROBUST: IBM Employee Use Case
Business Data Created per day Number of users

2007 2009 2013 2007 2009 2013

IBM Activities Entry 700 2750 5000 53200 143600 200000

IBM Blogs Entries 120 30 60 34600 77750 100000

IBM Communities 3 23 50 3000 181950 250000

IBM Bookmarks 800 900 1000 8500 22400 50000

IBM Wikis NA 40 100 NA 35450 100000

IBM Files NA 290 1000 NA 45160 100000
IBM Overall 1623 4033 7210 500000* 500000* 500000*


Risks in Online Communities
Definition: Risk Likelihood
 Probability of an event occurring
 Impact of the event occurring

Risk management Cost Benefit
 Process for managing costs, benefits and likelyhoods
 Detect high impact risks in time even if
they generate expensive false alarms SAP: SCN Award Points Scamming
 Ignore very low impact risks • Experts reputation decreases
even if they can be reliably detected • Business users leave the forum

Types of risks
 Non-compliance with the community policies/polity
 Scamming or spamming behavior
 Lower involvement and productivity
 Decrease of user satisfaction
 Loss of community dynamics
Web: Public communities
• Death of TechCrunch forum due to
Loss of 1% experts  loss of high revenue
spam and lack of management
Loss of 10% lurkers  low impact
8 staab@uni-koblenz.de Summer School 9

Communities: dynamics and confidentiality

ROBUST supports decision making for users, hosts and service providers
Managing growth & decline
 Identify, encourage, safeguard core users
 Social matching
 Define/maintain etiquette and policies
 Manage negative behavior and conflicts
 Content matching
 Recognize, categorize decline and growth
 Redirect users to other communities
Merging communities
 Cross community topic detection to stimulate inter-community interactions
Splitting communities
 Identification of clusters/compartments of members that can be separate


Agenda

• Risks and Opportunities in Social Communities:
the ROBUST project

• Many related Talks in this Summer School

Robust partners Closely related
Alani: Monitoring and analysis Greene: Network Analysis
of social networks Bernstein: Scalable
Karnstedt: User churn infrastructures

But here comes the biased account from work in our institute


Plan for this Talk

1 Web

2 Science


Bild eines schwarzen Lochs

staab@uni-koblenz.de Flickr cc, Jan 7 2009 by
Summer School 13 thebadastronomer

Agenda

the ROBUST project

• Web Science Methodology:
An explanation by analogy with Physics
and some initial (!) applications to online communities
• Modeling dynamic system at micro level,
Understanding collective effects (macro level) arising
from individual behavior (micro level)
• Predicting dynamic system behavior,
recognizing behavior deviating from the model
• Modeling dynamic system behavior at the macro level
• Controling dynamic system behavior by collective action

Better understanding of the tagging process
 Cooperative classification of resources
 Which factors influence the tagging process?
• Background knowledge of the user?
• Tag assignments of other users?

 Hypothesis: Tagging involves imitation of other users AND
selection of tags from background knowledge of users.

Methodology
User interface Something else?

Tagging
Conceptualization Behavior

Comparison
of Statistics
Own Shared
Knowledge terminology

Model of User Interface Influence

Simulated
Joint Stochastic Model Tagging
Behavior

Model of Own Model of
Knowledge Sharing

Components of Analysis

Properties of Tag Streams
Observations
 Stream view of Folksonomies in
 Co-occurrence streams the real world
 Resource streams

Dynamic model for Tagging Systems Stochastic
 Simulating background knowledge models of
 Simulating tag imitation influence

Simulation Results Which models
 Co-occurrence streams best fit the
 Resource streams reality?


Stream Views of a Folksonomy

Folksonomies:
 Vertices: Users, tags, resources
 Edges: Tag assignments
 Postings:
• Tag assignments of a user to a single resource
• Can be ordered according to their time-stamp


Co-occurrence Streams

Co-occurrence Streams:
 All tags co-occurring with a given tag in a posting
 Ordered by posting time
Co-occurrence stream for 'apple':
 {mackz, r1, {apple, tree}, 13:25}
{klaasd, r2, {apple, mac, ibook}, 13:26}
{mackz, r2, {apple, macintosh, stevejobs}, 13:27}
 tree, mac, ibook, macintosh, stevejobs

Tag |Y| |U| |T| |R|
ajax 2.949.614 88.526 41.898 71.525
blog 6.098.471 158.578 186.043 557.017
xml 974.866 44.326 31.998 61.843


Properties of Co-occurrence Streams – Tag Growth

linear
growth


Properties of Co-occurrence Streams – Tag Frequencies

power law


Resource Streams

Resource Streams:
 All tags assigned to a resource
 Ordered by posting time
Resource stream for 'r2':
 {mackz, r1, {apple, tree}, 13:25}
{klaasd, r2, {apple, mac, ibook}, 13:26}
{mackz, r2, {apple, macintosh, stevejobs}, 13:27}
 apple, mac, ibook, apple, macintosh, stevejobs


Properties of Resource Streams – Tag Frequencies



Simulating the Evolution of Tag Streams

Simulating tag streams
Which of my concepts
Inspiration for conceptualization from:
represent this web
page? How do I tag 1. Most popular tags
this web page?
2. Most recently used tags
3. Tags used for this resource
4. Tags co-occuring with similar text
documents
5. Creating completely new tags
6. …

Which combination of
inspirations develop the
same statistics as the
one observed for
delicious?


The Delicious User Interface

Imitating previous tag assignments:
 Recommended tags: Intersection of tags of a user and tags already
assigned to the resource.
 Your tags: Tags of the user.
 Popular tags: 7 most popular tags assigned to the resource.


Simulating a Tag Stream

Start with empty tag stream
Each simulation step appends a new tag assignment
Simulation of a single tag assignment:

p(w|t): Probability of selecting word w for topic t.
Modeled by word distributions in a topic centered
text corpus.

n: Number of visible previous tags.
h: Maximal number of previous tag assignments
used for determining ranking of the n distinct tags.


Modeling Background Knowledge

Text Corpora Del.icio.us
Text Corpora

PBK: Probability of selecting from background knowledge
 p(w|t): Probability of selecting word w for topic t. Modeled by word
distributions in a topic centered text corpus.
 p(w|r): Probability of selecting word w for resource r.

Modeling Tag Imitation

PBK t t-1 t-2 t-3 t-4 t-5 … t-h …

1-PBK
1 2 3 … n

PI = 1 – PBK: Probability of imitating a previous tag assignment
 n: Number of visible top-ranked tags
 h: Maximal number of previous tag assignments used for determining
ranking of the n distinct tags



Simulation Results

Overall Scheme

Tagging

Comparison
of Statistics
Own Shared


Simulated
Behavior

Model of Own Knowledge Model of Sharing


Simulating Co-occurrence Streams

Tag growth:
 Influenced by PBK and p(w|t)

Tag Frequencies:
 Influenced by PBK, p(w|t), n, h
 n: Semantic breadth of a topic (blog: 100 tags,
ajax: 50 tags, xml: 50 tags; Cattuto et al. 2007)
 h: No hint for realistic values. Good guesses may be 500
and 1000.


Co-occ. Streams – Simulated Tag Growth


Co-occ. Stream – Simulated Tag Frequencies


Simulating Resource Streams

PI and PBK: Values comparable to co-occurrence streams
p(w|r): Approximated by p(w|t)
n: 7 tags are visible (cf. Delicious user interface)
h: Smaller value than for co-occurrence streams


Res. Streams – Simulated Tag Frequencies


Lessons learned [Dellschaft+Staab,
ACM Hypertext 2008]

Black holes do not only eat mass they also dissolve by
emitting radiation

Imitation AND background knowledge are needed for
explaining properties of tag streams
Probability of imitating previous tag assignments: ~70-90%

Frequency Rank
Co-occur. Streams Resource Streams Tag Growth
Polya Urn Model o o fixed size
Simon Model o o linear
YS Model w/ Memory + o linear
Halpin et al. Model o o linear
Our Model
Epistemic Model + + power-law


Solar System

Neptun

Uranus
Jupiter

Saturn
Flickr, cc Sep 1 2008 by Image Editor

Agenda

the ROBUST project

Understanding collective effects (macro level) arising from
individual behavior (micro level)

Overall Scheme

Tagging

Comparison
of Statistics
Own Shared


Simulated
Behavior

Model of Own Knowledge Model of Sharing


What is our Uranus?

What is this?


Uranus = Spam [Dellschaft+Staab,
WebSci 2010]
Effect of removing 257 spammers of 12.777 users from the ‘bookmark’ stream


Why care? The Bibsonomy Example

Complete snapshot of Bibsonomy system
Manually labeled ground truth of spammers in the data set

Users Tags Resources TAS
Spammers 29,248 297,846 1,197,354 13,258,759
Non-Spammers 2,467 61,154 234,143 816,196


Why care? The Delicious Example

Crawled during the TAGora Project
Users Tags Resources TAS
532,938 2,482,850 18,778,566 140,305,446

Amount of spammers not known exactly
Estimation based on random sample of 500 users:
 With 95% probability: Between 1.972 and 12.949 spammers
 Delicious most likely already applies spam detection
 Why care about ~ 1.5% spammers in Delicious?


Filtering Results (Users)

Number of Spammers and Non-Spammers
16000

14000

12000

10000

Spammer
8000 Non-Spammer

6000

4000

2000

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20


Filtering Results (Tag Assignments)

Filtered and unfiltered number of TAS
450000

400000

350000

300000

250000
Spam
Non-Spam
200000

150000

100000

50000

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20


That’s why

Effect of removing 257 spammers of 12.777 users from the ‘bookmark’ stream


How statistically significant is the epistemic model for
normal users?


Lessons learned

Uranus was discovered because it affected Neptun
Pluto was discovered because it affected Uranus!

Spammers can be discovered by their behavior,
even if you do not know what kind of spam they are
producing!


How do constellations in the sky evolve?

http://www.flickr.com/photos/furious-angel/2142647358/sizes/o/in/photostream/

Agenda

the ROBUST project


Example: Network

Person Friendship


SUGGESTING WHOM TO LINK
TO NEXT


Use Networks for Recommendation

:-(

me

 Goal: Predict who a person will add as friend

 Facebook's algorithm: find friends-of-friends

→ Problem: Rest of the network is ignored!


Algebraic Graph Theory

3

1 2 4 5 6

Represent a network 1 2 3 4 5 6
1 0 1 0 0 0 0
by an adjacency matrix A:
2 1 0 1 1 0 0
3 0 1 0 1 0 0
Aij = 1 when i and j are connected A=
4 0 1 1 0 1 0
Aij = 0 when i and j are not connected 5 0 0 0 1 0 1
6 0 0 0 0 1 0
A is square and symmetric.


Baseline: Friend of a Friend Model

Count the number of ways a person can be found as
the friend of a friend.

Consider the matrix product AA = A2

2 3
0 1 0 0 0 0 1 0 1 1 0 0
1 0 1 1 0 0 0 3 1 1 1 0
0 1 0 1 0 0 1 1 2 1 1 0
=
0 1 1 0 1 0 1 1 1 3 0 1
0 0 0 1 0 1 0 1 1 0 2 0 1 2 4
0 0 0 0 1 0 0 0 0 1 0 1


Eigenvalue Decomposition

Write the matrix A as a product:

A = UΛUT

where
U are the eigenvectors UTU = I
Λ are the eigenvalues Λij = 0 when i ≠ j


Computing A2

Use the eigenvalue decomposition A = UΛUT

A2 = UΛUT UΛUT = UΛ2UT
Exploit U and Λ:
T
 U U = I because U contains eigenvectors
 (Λ ) = Λ because Λ contains eigenvalues
2 2
ii ii

Result: Just square all eigenvalues!


Friend of a Friend of a Friend
3

1 2 4 5 6

Compute the number of friends-of-friends-of-friends:
1 2 3 4 5 6
3
0 1 0 0 0 0 0 3 1 1 1 0 1
1 0 1 1 0 0 3 2 4 5 1 1 2
0 1 0 1 0 0 1 4 2 4 1 1 3
0 1 1 0 1 0 = 1 5 4 2 4 0 4
0 0 0 1 0 1 1 1 1 4 0 2 5
0 0 0 0 1 0 0 1 1 0 2 0 6

A3 = UΛUT UΛUT UΛUT = UΛ3UT

Matrix Exponential
3

0.98 0.76 0.22
1 2 4 5 6 7

The matrix exponential can be written as a power
sum with decreasing coefficients:
exp(A) = I + A + 1/2 A2 + 1/6 A3 + . . .
1 2 3 4 5 6 7
0 1 0 0 0 0 0 1.66 1.72 0.93 0.98 0.28 0.06 0.01 1
1 0 1 1 0 0 0 1.72 3.57 2.70 2.93 1.04 0.29 0.06 2
0 1 0 1 0 0 0 0.93 2.70 2.86 2.71 0.99 0.28 0.06 3
exp 0 1 1 0 1 0 0 = 0.98 2.93 2.71 3.63 1.95 0.76 0.22 4
0 0 0 1 0 1 0 0.28 1.04 0.99 1.95 2.35 1.59 0.64 5
0 0 0 0 1 0 1 0.06 0.29 0.28 0.76 1.59 2.23 1.38 6
0 0 0 0 0 1 0 0.01 0.06 0.06 0.22 0.64 1.38 1.59 7

Recommendations for user ④: ①>⑥>⑦

Why the Matrix Exponential

An
= Number of paths of length n

aA2 + bA3 + cA4 + . . .
= Number of paths, weighted by path length

→ New edges more likely to appear when there are
many paths already

→ When a > b > c > . . . > 0, short paths are
weighted more


Computing Power Series

Let p(A) be a power series:

p(A) = aA2 + bA3 + cA4 + . . .
= aUΛ2UT + bUΛ3UT + cUΛ4UT + . . .
= U(aΛ2 + bΛ3 + cΛ4 + . . .)UT
= Up(Λ)UT

Therefore:

Power series change only the eigenvalues!


TRACKING THE EVOLUTION
OF THE NETWORK AS A
WHOLE

Diversity
• Many, equally-sized subcommunities
• High entropy
• ‘Flat’ structure

Regularity
• Few large subcommunities
• Low entropy
• Many ‘hubs’


⇒ ⇒
Network Evolution

• How did a network look at time t?
• Idea: Observe the change of diversity/regularity over time


Outline

1. Power-law exponent
2. Weighted spectral distribution
3. Network entropy
4. Network rank


1. Power-law Exponent

Number of neighbors is unevenly distributed:

Epinions trust network (Massa et al. 2005)

C(n) ∼ n−γ

Results in a power-law (Newman 2006)
Higher exponent γ denotes less regularity

1. Power-law Exponent over Time

Epinions trust network (Massa et al. 2005)

γ shrinks ⇒ Network becomes more regular

2. Weighted Spectral Distribution

• Consider the n×n matrix N defined by

Nij = 1 / sqrt(d(i)d(j)) when (i,j) is an edge
Nij = 0 otherwise

Then the distribution of the eigenvalues of N is called the
weighted spectral distribution (WSD) (Fay et al. 2010)

Eigenvalues nearer to ±1: diversity
Eigenvalues nearer to 0: regularity


2. Weighted Spectral Distribution over Time

CiteULike user–tag network (Emamy et al. 2007)

• The WSD shifts to zero ⇒ Regularization
The WSD shifts towards zero ⇒ The network becomes regular

3. Network Entropy

G = G1 ∪ G2 ∪ . . . ∪ Gr
• Write the graph G as a sum of subgraphs Gk

Each Gk has weighted edges, with total weight λk
• When picking an edge from G at random, the probability of
it being in community Gk is
λk / (λ1 + λ2 + . . . + λr) = λk / L
• The entropy of this distribution is (Kunegis et al. 2011)

H(G) = − Σk (λk / L) log (λk / L)
• Entropy: Effective number of subcommunities


3. Network Entropy over Time

Enron email network (Klimt et al. 2004)
absolut
e
Entropy (H(G))

zoo
m

Entropy is constant ⇒ Constant number of communities
0
Time (t)


4. Network Rank

Decompose network into subcommunities:

G = G1 ∪ G2 ∪ . . . ∪ Gr

The rank r is a measure of diversity:

rank(G) = r

Weighted rank:

rank∗(G) = Σk |Gk| / |G1|
Robust measure of diversity (Kunegis et al. 2011)

4. Network Rank over Time

Network rank (rank∗(G))

Enron email network (Klimt et al. 2004)

Time (t)

• Increasing network rank: increasing diversity
• Shrinking network rank: shrinking diversity

More Network Rank Plots

Epinions trust network
hep-th citations Wikipedia elections

frwikibooks edits MIT conference contacts YouTube social network

(biased towards good examples of convex evolution)

Conclusion

• Power-law exponent shrinks
– Connection diversity shrinking
• Weighted spectral distribution shifts to zero
– Emerging main components
• Entropy is constant
– Effective number of communities is constant
• Network rank increases, then shrinks
– Two-phase- model of expansion


Watch out!

KONECT – Koblenz Network Collection

http://uni-koblenz.de/~kunegis/paper/kunegis-
konect.poster.pdf

Coming soon!

Follow #ictrobust or @kunegis or @ststaab


Why has the sky the density it has?

staab@uni-koblenz.de Summer School 81 14,
Flickr, cc Oct 2007, Michael Donough

Why do tagging systems have so little spam?

Administrative
Process

Content Community User
Quality Policy Roles

Content
Steffen Staab
Process
Web Science Doctoral

Agenda

the ROBUST project


Yahoo Answers

• Ensure quality of user generated content

• Use of administrators and community moderators How?

• Policy influences community processes


SURVEY OF
GOVERNANCE MODELS


Communities need Governance

 Steering and coordinating actions of community members
[Benz2004]
Goal: Successful and flourishing community
 High quality user-generated content
 Active community members
[ http://www.flickr.com/photos/61433480@N02/5593890914/, http://www.flickr.com/photos/boojee/3733902852/ ]


Motivation

Different types of
 Web communities
 User-generated content (video, photos, comment, article,
questions, answers, posting, review text)

What are the most successful means of
governance for user-generated content?

Analyze successful platforms and compare
their means of governance!


Means of Governance

1. Direct intervention of community owner
Affecting content or users based on apparent properties
2. Functionality of the community platform

Text Reviews Bookmarks
Ratings Abuse Reports
Assessment

User-generated Content Modification Community
Content Complex User Roles Member
Selection & Ranking
Ratings Score
Time Views Replies Hide Low Quality


Method

Selection of 250 most prominent web sites with community
functionality according to Alexa Page Rank
Clustering web sites in four groups according to purpose
Social Media Editorial News

Social Networking Social Reviewing

Top-5 web sites of each group analyzed (*)

Key Results

(1) Abuse Reports are a successful means of governance.
• 16 occurrences
• Restricted to filter out unwanted content
• Staff needed – expensive but efficient [Schwagereit2010]

(2) Simple ratings are dominant – but battle between
“Like” and “Like/Dislike”
• “Like”: 9 occurrences
• “Like/Dislike”: 7 occurrences
• Tradeoff between simplicity and improved ranking ability


Key Results

(3) Creation time is most implemented ranking criterion
• 18 occurrences
• Others: score: 8, ratings: 6
• Important content is renewed - unimportant content will be
forgotten

(4) Content modification and user roles are rarely
implemented
 2 occurrences
 Requires complex role system and users
who understand it


GOVERNANCE MODEL:
DEEP DIVE - SIMULATION


Methodology Principle

1. Define a Web Community model
(Lycos IQ, Yahoo Answers…)
2. Adapt this model to an existing community
3. Estimate parameters

4. Define quality measure

5. Simulate community behaviour

6. Compare simulation results with real data
7. Analyze quality measures wrt variations of CoSiMo
parameters


Dataset Lycos IQ

Time Period: 909 days
Users: 34.327
Administrators: 36
Questions: 1.031.982
Answers: 2.996.446
Deleted non-compliant Answers: 21.139


Observed parameters (input to simulation)

100000

10000

1000

100 Number of
Users
10

1
0-999
1000-1999
2000-2999
3000-3999
4000-4999
5000-5999
0.9-1.0
0.8-0.89

6000-6999
0.7-0.79
0.6-0.69

7000-7999
0.5-0.59
0.4-0.49

>7000
0.3-0.39
0.2-0.29

Answers
0.1-0.19
0.0-0.09

per year
Rate of Compliant Answers

Example Behaviors and Example Policies

Behaviors of Ordinary Users: Reading Policies for
• Create new postings Administrators:
• Read existing postings PA: random selection of
• Report non-compliant postings
postings PB: random selection of
OR give bonus points to postings that no other
poster administrator has examined
so far
Moderator Users: PC: selection of postings that
• Create new postings were most often reported
• Read existing postings by users for being non-
• Delete non-compliant
compliant
posting
OR give bonus points to Promotion Policy:
poster PM-X : ordinary users become
moderators (who can
Administrators: delete postings) when
•Read existing postings having at least X bonus
•Delete non-compliant points
postings

How many administrators are needed?

1,05 0,95-
1,05
0,95 0,85-
Recent 0,95
0,85 Posting 0,75-
Quality 0,85
0,75
0,65-
0,65 0,75

5
10
20
40
1152

80
288
72 160 Additional non-compliant
320
18
640

4 Postings (per day)
1280

Number of Administrators
2560

1


Fighting spam with administrators…

1
0,998
0,996
0,994 0,998-1
Recent 0,992
Posting 0,99 0,996-0,998

576
Quality 0,994-0,996

72
9
0,992-0,994

1
0,99-0,992
Number of
Administrators
Applied Policies

Variation of policies and number of administrators
• Efficient policies result in high quality content
• A minimum of 18 administrators are needed
• Many moderators are needed to bring the quality to a high level

Fighting spam with user moderators…

1
0,95
0,9
0,85
0,8 0,95-1
0,75
0,7 Recent
5 0,65
100,6 Posting
20
40 0,9-
80
160 Quality
320
640 0,95
1280
PA+PB+PC+PM12

2560 0,85-
PA+PB+PC+PM25

PA+PB+PC+PM1…
PA+PB+PC+PM50

PA+PB+PC+PM3…
PA+PB+PC+PM100
PA+PB+PC+PM200
PA+PB+PC+PM400
PA+PB+PC+PM800

PA+PB+PC
PA+PB
0,9

PA
Additional non-
compliant
Postings (per
day)
Applied Policies

Variation of policies and posting quality
• A limited number of administrators has a limited capacity of
filtering a surge of non-compliant postings
• Moderators are helping to increase quality

Lessons Learned

• Strategy of selecting questionable postings is crucial

• Reporting by normal users is the most effective strategy

• Moderators are not so effective as expected, if they hunt
only incidentally for non-compliant content

• Sufficiently strong requirements regarding moderator
profiles lead to high quality of moderators

• Policies for promoting users need to be based on a
criterion that is time dependent


Agenda

the ROBUST project


Are we satisfied here? No! Not by far!

Understand how and why users tag or tweet?
-> What are people‘s limitations that affect the system?
-> Psychology and Sociology!

What are their legal boundaries?
-> How can you shape the systems?
-> Law!

What are organizations‘ incentives?
-> Why and how do organizations participate?
-> Nice example: open source
-> Economy


Thank You!

References

The Slashdot Zoo: Mining a social network with negative edges
J. Kunegis, A. Lommatzsch and C. Bauckhage
In Proc. World Wide Web Conf., pp. 741–750, 2009.
Learning spectral graph transformations for link prediction
J. Kunegis and A. Lommatzsch
In Proc. Int. Conf. on Machine Learning, pp. 561–568, 2009.
Spectral analysis of signed graphs for clustering, prediction and
visualization
J. Kunegis, S. Schmidt, A. Lommatzsch and J. Lerner
In Proc. SIAM Int. Conf. on Data Mining, pp. 559–570, 2010.
Network growth and the spectral evolution model
J. Kunegis, D. Fay and C. Bauckhage
In Proc. Conf. on Information and Knowledge Management,
pp. 739–748, 2010.


References

B. Viswanath, A. Mislove, M. Cha, K. P. Gummadi, On the
evolution of user interaction in Facebook. In Proc.
Workshop on Online Social Networks, pp. 37–42, 2009.


References

K. Dellschaft, S. Staab. An Epistemic Dynamic Model for
Tagging Systems. HYPERTEXT 2008, Proceedings of the
19th ACM Conference on Hypertext and Hypermedia,
June 19-21, 2008 - Pittsburgh, Pennsylvania, USA.
K. Dellschaft, S. Staab. On Differences in the Tagging
Behavior of Spammers and Regular Users. In: Proc. of
WebSci-2010, Raleigh, April, 2010.
F. Schwagereit, S. Sizov, S. Staab. Finding Optimal Policies
for Online Communities with CoSiMo. In: Proc. of WebSci-
2010, Raleigh, US, April, 2010.


Managing Social Communities

Empfohlen

Empfohlen

Weitere ähnliche Inhalte

Ähnlich wie Managing Social Communities

Ähnlich wie Managing Social Communities (20)

Mehr von Steffen Staab

Mehr von Steffen Staab (20)

Kürzlich hochgeladen

Kürzlich hochgeladen (20)

Managing Social Communities