Viability of Online Communities -...

Viable online communities Wiki dynamics Flickr group dynamics Research directions

Viability of Online CommunitiesTwo case studies

Dario Taraborelli* Camille Roth†

* Centre for Research in Social SimulationUniversity of Surrey

† Centre d’Analyse et de Mathematique SocialesCNRS-EHESS

ECCS 2009PATRES Satellite Meeting

September 24, 2009Viability of Online Communities D. Taraborelli, C. Roth


Sustaining the viability of online communities

Typical risks faced by online communities:I content explosionI active population extinctionI high user turnoverI high user drop-off rate

Typical solution:→ best practices

Typical shortcomings:I Poor self-awareness of policy and

governance effects on communitydynamics

I Lack of tools forempirically-grounded predictions

“Wikipatterns is a toolbox of patternsand anti-patterns. Looking to spur wikiadoption? Want to grow from 10 usersto 100, or 1000? Applying patternsthat help guide the growth of content,and recognizing anti-patterns thatmight hinder growth, can give yourwiki the greatest chance of success”

Viability of Online Communities D. Taraborelli, C. Roth


Overview

1. Case study: Peer production systems

I Growth enhancers and regulators in wiki-basedcommunities

2. Case study: Social media

I Group affiliation dynamics in Flickr groups

3. Directions

I Constraints on models of viable online communities



Peer production systems: the case of wikis

Collaborative structure of a wiki:

I populationI editors (contributing

content)I administrators

(managing edit revisions,

user access restrictions)

I contentI individual editsI collaborative edited

content (pages)

I governanceI access restriction

(determine who can edit

what)



Wiki dynamics: data source

I Data sourcePopulation and content evolution tracked over an 8-monthspan for a sample of 11,500+ MediaWikis

I Selected sample360 wikis, with an initial population between 400 and20,000 users, restricted to hosters with reliable data, andwith controlled discontinuities in daily change rates.

I VariablesI population size (U)I content size (P)I admin population (A)I edits (E)I access control (R)

1 10 100 1000 10000 100000

1

10

100

1000

10000

100000

1000000

1 10 100 1000 10000 100000

1

10

100

1000

10000

100000

1000000



Metrics considered

demographic metrics

I user activity, i.e. the proportion of edits per user (E/U)I user density, i.e. the proportion of users per page (U/P)I edit density, i.e. the proportion of edits per page (E/P).

governance metrics

I administrator ratio,i.e. the proportion of users with admin status (A/U)

I administrator density,i.e. the proportion of admins per page (A/P)

I editing permission (R).

Governance-based growth drivers can be controlled!Viability of Online Communities D. Taraborelli, C. Roth


Assessing the effects of wiki metrics on growth

1. Sort wikis according to each of the above metrics (e.g. editdensity)

edit density

2. Aggregate wikis in bins (quantiles) of identical size (e.g.wikis with low edit density vs. wikis with high edit density)

Q1 Q2 Q3 Q4 Q5

edit density

3. Measure how each quantile performs with respect tocontent and population growth rates

Q1 Q2 Q3 Q4 Q5

edit density



Growth enhancers (1): user activity

of their population or content.

3. WIKI DYNAMICSWe assessed wiki dynamics by comparing their diverse paths withrespect to a set of independent variables. ‘Growth’ is defined interms of population and content size change: user growth GU (resp.page growth GP ) is the ratio between final and initial populations(resp. content sizes): GU = Ulast/Ufirst (resp. GP = Plast/Pfirst).Wiki dynamics were studied as a function of the variables listed inSection 2.2:

(I) DESCRIPTIVE INDICATORS, i.e. variables on which wiki ad-ministrators have no direct control: (a) user activity, i.e. theproportion of edits per user (E/U ), (b) user density, i.e. theproportion of users per page (U/P ), and (c) edit density, i.e.the proportion of edits per page (E/P ).

(II) GOVERNANCE FACTORS, variables that wiki administratorscan directly control: (a) administrator ratio, i.e. the propor-tion of users who are granted administrator status (A/U ),(b) administrator density, i.e. the proportion of administra-tors per page (A/P ), (c) editing permission (R).

For each continuous variable, instead of carrying out a delicateanalysis by dealing with clouds of points, we adopted a more in-sightful approach by dividing wikis into five quantiles, each includ-ing exactly 20% of all wikis in the clean dataset (see Table 1). We

1 2 3 4 5

Quantile of

edits per user1.0

1.1

1.2

1.3

1.4

1.5

1.6

Growth ratio

User growth

1 2 3 4 5

Quantile of

edits per user

0.95

1.00

1.05

1.10

1.15

1.20

1.25

1.30

Growth ratio

Page growth

1.1

1.11.1

1.2

1.2

1.3

1.3

1.4

1.4

1 2 3 4 51

2

3

4

Quantiles of 'edits per user'

Populationquantiles

User growth

1

1

1

1.1

1.1

1.1

1.1

1.11.2

1.2

1 2 3 4 51

2

3

4


Populationquantiles

Page growth

Figure 1: Growth landscape with respect to the proportion ofedits per user.

1 2 3 4 5

Quantile of

users per good page1.0

1.1

1.2

1.3

1.4

1.5

1.6

Growth ratio

User growth

1 2 3 4 5

Quantile of


1.05

1.10

1.15

1.20

1.25

1.30

Growth ratio

Page growth

1

1.1

1.1

1.2

1.2

1.3

1.3

1.41.4 1.5

1 2 3 4 51

2

3

4

Quantiles of 'users per good page'

Populationquantiles

User growth

1

1

1

1.1

1.1

1.1

1.1

1.2

1 2 3 4 51

2

3

4


Populationquantiles

Page growth

Figure 2: Growth landscape with respect to the proportion ofusers per good page.

then computed and compared growth ratio means over all wikis foreach quantile. Additionally, we distinguished population quantilesin order to control for user size-related effects. To this end, weplotted a growth landscape that consists of a two-dimensional rep-resentation of the various growth ratios. This representation wasapplied to all the above-mentioned variables, except for R wherethere are only two “quantiles” (0 or 1). For each variable except R,the upper graphs indicate the mean values and confidence intervals(p < 0.05) of each quantile on the variable considered, while thelower graphs show contour plots for the same variable with brighterareas corresponding to higher growth ratios.

3.1 Significant descriptive indicatorsWe found significant correlations between a number of descriptiveindicators of wiki structure and their content and population growthrates.

Figure 1 shows the effect of user activity (measured as the pro-portion of edits per user) on growth rates. The results suggest thatuser activity correlates very strongly with wiki growth, not only interms of content production (which is to a certain extent unsurpris-ing) but also new member recruitment. The effect becomes strongerwith initially more populated wikis: the more users are activelyediting, the more a wiki grows in content and population.

Figure 2 shows the impact of user density on growth. The re-sults suggest that a higher number of contributors per page does notnecessarily indicate mushrooming wikis: for an identical contentsize, we found a significant correlation between a lower number ofusers and higher growth ratios, both in content and new members.

To better visualize the effect of user density on growth, we rep-resented the dependent variables GU and GP , independent vari-

1

2

3

4

5

6

7

8

U!P

400 1000 10000 20000

10

100

1000

10000

100000

Population !users"

Content

!goodpages"

Figure 3: Phase diagram in the content/population space ofwikis belonging to the clean dataset.



Growth regulators (1): user density

of their population or content.

3. WIKI DYNAMICSWe assessed wiki dynamics by comparing their diverse paths withrespect to a set of independent variables. ‘Growth’ is defined interms of population and content size change: user growth GU (resp.page growth GP ) is the ratio between final and initial populations(resp. content sizes): GU = Ulast/Ufirst (resp. GP = Plast/Pfirst).Wiki dynamics were studied as a function of the variables listed inSection 2.2:

(I) DESCRIPTIVE INDICATORS, i.e. variables on which wiki ad-ministrators have no direct control: (a) user activity, i.e. theproportion of edits per user (E/U ), (b) user density, i.e. theproportion of users per page (U/P ), and (c) edit density, i.e.the proportion of edits per page (E/P ).

(II) GOVERNANCE FACTORS, variables that wiki administratorscan directly control: (a) administrator ratio, i.e. the propor-tion of users who are granted administrator status (A/U ),(b) administrator density, i.e. the proportion of administra-tors per page (A/P ), (c) editing permission (R).

For each continuous variable, instead of carrying out a delicateanalysis by dealing with clouds of points, we adopted a more in-sightful approach by dividing wikis into five quantiles, each includ-ing exactly 20% of all wikis in the clean dataset (see Table 1). We

1 2 3 4 5

Quantile of

edits per user1.0

1.1

1.2

1.3

1.4

1.5

1.6

Growth ratio

User growth

1 2 3 4 5

Quantile of

edits per user

0.95

1.00

1.05

1.10

1.15

1.20

1.25

1.30

Growth ratio

Page growth

1.1

1.11.1

1.2

1.2

1.3

1.3

1.4

1.4

1 2 3 4 51

2

3

4


Populationquantiles

User growth

1

1

1

1.1

1.1

1.1

1.1

1.11.2

1.2

1 2 3 4 51

2

3

4


Populationquantiles

Page growth

Figure 1: Growth landscape with respect to the proportion ofedits per user.

1 2 3 4 5

Quantile of


1.1

1.2

1.3

1.4

1.5

1.6

Growth ratio

User growth

1 2 3 4 5

Quantile of


1.05

1.10

1.15

1.20

1.25

1.30

Growth ratio

Page growth

1

1.1

1.1

1.2

1.2

1.3

1.3

1.41.4 1.5

1 2 3 4 51

2

3

4


Populationquantiles

User growth

1

1

1

1.1

1.1

1.1

1.1

1.2

1 2 3 4 51

2

3

4


Populationquantiles

Page growth

Figure 2: Growth landscape with respect to the proportion ofusers per good page.

then computed and compared growth ratio means over all wikis foreach quantile. Additionally, we distinguished population quantilesin order to control for user size-related effects. To this end, weplotted a growth landscape that consists of a two-dimensional rep-resentation of the various growth ratios. This representation wasapplied to all the above-mentioned variables, except for R wherethere are only two “quantiles” (0 or 1). For each variable except R,the upper graphs indicate the mean values and confidence intervals(p < 0.05) of each quantile on the variable considered, while thelower graphs show contour plots for the same variable with brighterareas corresponding to higher growth ratios.

3.1 Significant descriptive indicatorsWe found significant correlations between a number of descriptiveindicators of wiki structure and their content and population growthrates.

Figure 1 shows the effect of user activity (measured as the pro-portion of edits per user) on growth rates. The results suggest thatuser activity correlates very strongly with wiki growth, not only interms of content production (which is to a certain extent unsurpris-ing) but also new member recruitment. The effect becomes strongerwith initially more populated wikis: the more users are activelyediting, the more a wiki grows in content and population.

Figure 2 shows the impact of user density on growth. The re-sults suggest that a higher number of contributors per page does notnecessarily indicate mushrooming wikis: for an identical contentsize, we found a significant correlation between a lower number ofusers and higher growth ratios, both in content and new members.

To better visualize the effect of user density on growth, we rep-resented the dependent variables GU and GP , independent vari-

1

2

3

4

5

6

7

8

U!P

400 1000 10000 20000

10

100

1000

10000

100000

Population !users"

Content

!goodpages"

Figure 3: Phase diagram in the content/population space ofwikis belonging to the clean dataset.Viability of Online Communities D. Taraborelli, C. Roth


Growth enhancers (2): edit permission

The benefits of openness:unconstrained access vs. edit restrictions

able U/P , and initial wiki positions Ufirst and Pfirst altogether onthe same graph, yielding a phase diagram as plotted on Fig. 3.6 Inthis diagram, each dot (light color) corresponds to a wiki in thedatabase. Each arrow corresponds to a pair of quantiles “users perpage, population”. Widths and heights are proportional to user andcontent growth ratios, respectively. The size of the arrow representsthe strength of the observed growth in content and population forwikis in a given region of the wikisphere.

This graph should be regarded as a map of a portion of the wiki-sphere, showing the expected destiny of a wiki in terms of contentand population growth as a function of its initial position in thesame space. This diagram broadly suggests that a wiki’s positionis correlated with its subsequent fate. More precisely, it illustratesthat wikis in the upper/upper-right portion of the diagram are grow-ing faster, and more interestingly it provides an overview of demo-graphic dynamics in this region of the wikisphere.

3.2 Significant governance factorsTurning to governance features, we first analyzed the effects of theadministrator density on wiki dynamics by looking at the overallproportion of administrators per page.

Figure 4 shows that having a relatively high number of adminis-trators for a given content size is likely to reduce growth. There is astrong effect of the proportion of admins per page both on user andpage growth. For instance, while the last quantile of admins/pageratio enjoys near-zero growth rates over 8 months, the first quantiletops overall rates (!+50% for users, !+25% for pages). This ef-fect may be interpreted as the impact of strong governance activityon the proliferation of content and users.

6For this diagram, an increased level of detail called for a larger grid, here of 8 ! 7quantiles; U/P quantile means are represented by diagonal straight lines labelled “1–8”.

1 2 3 4 5

Quantile of

admins per good page1.0

1.1

1.2

1.3

1.4

1.5

1.6

Growth ratio

User growth

1 2 3 4 5

Quantile of


1.00

1.05

1.10

1.15

1.20

1.25

1.30

Growth ratio

Page growth

1.1

1.1

1.2

1.3

1.3

1.4

1 2 3 4 51

2

3

4

Quantiles of 'admins per good page'

Populationquantiles

User growth

1

1

1

1

1.1

1.1

1.1

1.1

1.2

1 2 3 4 51

2

3

4


Populationquantiles

Page growth

Figure 4: Growth landscape with respect to the proportion ofadmins per good page.

1 2 3 41.0

1.1

1.2

1.3

1.4

1.5

Population quantiles

Growthrate

User growth

1 2 3 41.00

1.05

1.10

1.15

1.20

1.25

1.30

1.35

1.40


Growthrate

Page growth

Figure 5: Growth landscape with respect to editing permission:red dashed refers to anonymously editable wikis, while bluesolid to wikis editable by registered users only.

1 2 3 4 5

Quantile of

edits per good page1.0

1.1

1.2

1.3

1.4

1.5

Growth ratio

User growth

1 2 3 4 5

Quantile of


1.05

1.10

1.15

1.20

1.25

Growth ratio

Page growth

1.2

1.2

1.2

1.3

1.3 1.4

1 2 3 4 51

2

3

4

Quantiles of 'edits per good page'

Populationquantiles

User growth

1.1

1.1

1.1

1.1

1.1

1.1

1 2 3 4 51

2

3

4


Populationquantiles

Page growth

Figure 6: Growth landscape with respect to the proportion ofedits per good page.

1 2 3 4 5

Quantile of

admins per user1.0

1.1

1.2

1.3

1.4

1.5

Growth ratio

User growth

1 2 3 4 5

Quantile of

admins per user1.00

1.05

1.10

1.15

1.20

1.25

Growth ratio

Page growth

1.2

1.2

1.3

1.3 1.4

1 2 3 4 51

2

3

4

Quantiles of 'admins per user'

Populationquantiles

User growth

1 11

1.1

1.1

1.1

1.1

1.1

1.1

1.1

1.1

1.2

1.2

1.2

1 2 3 4 51

2

3

4


Populationquantiles

Page growth

Figure 7: Growth landscape with respect to the proportion ofadmins per user.

We identified another significant effect when we considered edit-ing permission. As a binary variable, the editing permission vari-able generates only two groups of wikis (wikis that allow anony-mous editing versus wikis that restrict editing to registered usersonly). The growth landscape is consequently limited to a one-dimensional comparison over population quantiles. The results inFigure 5 show that for both dimensions—population and content—having no access control is likely to favor growth. While a strongerpage growth is quite unsurprising in wikis where no registrationis required, the fact that this factor also fuels user registration ismore puzzling. One might expect that if users can participate with-out the need of registration, few would be inclined to register. Ourresults suggest on the contrary that wikis with unrestricted registra-tion trigger participation more easily than wikis that restrict access.

3.3 Neutral indicatorsFinally, we consider two indicators that showed a markedly mildercorrelation with wiki dynamics.

On the one hand, we found that edit density (i.e. edits/page) cor-relates in a moderately negative way with user growth—with a rel-atively stronger effect depending on initial population size—whilethere is surprisingly no significant correlation with page growth(Figure 6).

On the other hand, higher administrator ratios (i.e. admins/user)have no significant effect on content or population growth, as evi-denced by the contour plot on Figure 7.



Growth regulators (2): admin density

able U/P , and initial wiki positions Ufirst and Pfirst altogether onthe same graph, yielding a phase diagram as plotted on Fig. 3.6 Inthis diagram, each dot (light color) corresponds to a wiki in thedatabase. Each arrow corresponds to a pair of quantiles “users perpage, population”. Widths and heights are proportional to user andcontent growth ratios, respectively. The size of the arrow representsthe strength of the observed growth in content and population forwikis in a given region of the wikisphere.

This graph should be regarded as a map of a portion of the wiki-sphere, showing the expected destiny of a wiki in terms of contentand population growth as a function of its initial position in thesame space. This diagram broadly suggests that a wiki’s positionis correlated with its subsequent fate. More precisely, it illustratesthat wikis in the upper/upper-right portion of the diagram are grow-ing faster, and more interestingly it provides an overview of demo-graphic dynamics in this region of the wikisphere.

3.2 Significant governance factorsTurning to governance features, we first analyzed the effects of theadministrator density on wiki dynamics by looking at the overallproportion of administrators per page.

Figure 4 shows that having a relatively high number of adminis-trators for a given content size is likely to reduce growth. There is astrong effect of the proportion of admins per page both on user andpage growth. For instance, while the last quantile of admins/pageratio enjoys near-zero growth rates over 8 months, the first quantiletops overall rates (!+50% for users, !+25% for pages). This ef-fect may be interpreted as the impact of strong governance activityon the proliferation of content and users.

6For this diagram, an increased level of detail called for a larger grid, here of 8 ! 7quantiles; U/P quantile means are represented by diagonal straight lines labelled “1–8”.

1 2 3 4 5

Quantile of


1.1

1.2

1.3

1.4

1.5

1.6

Growth ratio

User growth

1 2 3 4 5

Quantile of


1.00

1.05

1.10

1.15

1.20

1.25

1.30

Growth ratio

Page growth

1.1

1.1

1.2

1.3

1.3

1.4

1 2 3 4 51

2

3

4


Populationquantiles

User growth

1

1

1

1

1.1

1.1

1.1

1.1

1.2

1 2 3 4 51

2

3

4


Populationquantiles

Page growth

Figure 4: Growth landscape with respect to the proportion ofadmins per good page.

1 2 3 41.0

1.1

1.2

1.3

1.4

1.5


Growthrate

User growth

1 2 3 41.00

1.05

1.10

1.15

1.20

1.25

1.30

1.35

1.40


Growthrate

Page growth

Figure 5: Growth landscape with respect to editing permission:red dashed refers to anonymously editable wikis, while bluesolid to wikis editable by registered users only.

1 2 3 4 5

Quantile of


1.1

1.2

1.3

1.4

1.5

Growth ratio

User growth

1 2 3 4 5

Quantile of


1.05

1.10

1.15

1.20

1.25

Growth ratio

Page growth

1.2

1.2

1.2

1.3

1.3 1.4

1 2 3 4 51

2

3

4


Populationquantiles

User growth

1.1

1.1

1.1

1.1

1.1

1.1

1 2 3 4 51

2

3

4


Populationquantiles

Page growth

Figure 6: Growth landscape with respect to the proportion ofedits per good page.

1 2 3 4 5

Quantile of

admins per user1.0

1.1

1.2

1.3

1.4

1.5

Growth ratio

User growth

1 2 3 4 5

Quantile of

admins per user1.00

1.05

1.10

1.15

1.20

1.25

Growth ratio

Page growth

1.2

1.2

1.3

1.3 1.4

1 2 3 4 51

2

3

4


Populationquantiles

User growth

1 11

1.1

1.1

1.1

1.1

1.1

1.1

1.1

1.1

1.2

1.2

1.2

1 2 3 4 51

2

3

4


Populationquantiles

Page growth

Figure 7: Growth landscape with respect to the proportion ofadmins per user.

We identified another significant effect when we considered edit-ing permission. As a binary variable, the editing permission vari-able generates only two groups of wikis (wikis that allow anony-mous editing versus wikis that restrict editing to registered usersonly). The growth landscape is consequently limited to a one-dimensional comparison over population quantiles. The results inFigure 5 show that for both dimensions—population and content—having no access control is likely to favor growth. While a strongerpage growth is quite unsurprising in wikis where no registrationis required, the fact that this factor also fuels user registration ismore puzzling. One might expect that if users can participate with-out the need of registration, few would be inclined to register. Ourresults suggest on the contrary that wikis with unrestricted registra-tion trigger participation more easily than wikis that restrict access.

3.3 Neutral indicatorsFinally, we consider two indicators that showed a markedly mildercorrelation with wiki dynamics.

On the one hand, we found that edit density (i.e. edits/page) cor-relates in a moderately negative way with user growth—with a rel-atively stronger effect depending on initial population size—whilethere is surprisingly no significant correlation with page growth(Figure 6).

On the other hand, higher administrator ratios (i.e. admins/user)have no significant effect on content or population growth, as evi-denced by the contour plot on Figure 7.



Wiki growth landscape

1

2

3

4

5

6

7

8U!P

400 1000 10000 20000

10

100

1000

10000

100000

Population !users"

Content

!goodpages"



Summary of results

Editsper user

Editionpermission

Adminsper user

Edits pergood page

Users pergood page

Admins pergood page

0.9

1.0

1.1

1.2

Page growth

Editsper user

Editionpermission

Adminsper user

Edits pergood page

Users pergood page

Admins pergood page

0.7

0.8

0.9

1.0

1.1

1.2

1.3

User growth

C. Roth, D. Taraborelli, N. Gilbert (2008)Measuring wiki viability. An empirical assessment of the social dynamics of alarge sample of wikis. Proceedings of the 4th International Symposium onWikis (WikiSym 2008), New York, NY, USA, September 2008, ACM.



Social media: the case of Flickr groups

Collaborative structure of a Flickr group:

I populationI contributors (sharing photos with a group)I moderators (selecting submitted photos for

publication from the group moderation queue)I administrators (managing member rights)

I contentI photos shared by members with the

group (group pool)I group discussions

I governanceI privacy level (public/invite-only/private)I moderation queue (if enabled, photos must be selected by moderators to appear in the pool)I throttling (limits the max. number of photos users can post to the group per time period)



Flickr group dynamics: data source

Flickr Group Trackr: 15,000+ public Flickr groupsI Since Jan 2008: daily group variations in

population/content/governanceI Dataset restricted to 10K groups observed over 30 days



Flickr group dynamics: data source

Flickr Group Trackr: 15,000+ public Flickr groups



A glimpse at the Flickrverse

1 10 100 1000 104

1

100

104

106

Photos

Use

rs

I strong correlation between content and population growthI variety of individual group goals



Flickr groups growth landscape

1

234567

8

9

U=P

100 400 1000 10 000

10

100

1000

10 000

100 000

Population HusersL

Con

tent

Hpict

ures

L



Flickr groups vs. wikis growth landscape

1

234567

8

9

U=P

100 400 1000 10 000

10

100

1000

10 000

100 000

Population HusersL

Con

tent

Hpict

ures

L1

2

3

4

5

6

7

8U=P

400 1000 10 000 20 000

10

100

1000

10 000

100 000

Population HusersL

Con

tent

Hgood

page

sL



Peer production vs. social media

Macroscopic discrepancies

I demographic and governance properties are good predictors ofgrowth in the case of wikis

I ...but surprisingly poor at predicting growth rates in social mediacommunities.

Possible reasons

1. unnoticeable effects on relative growth rates as opposed to absolutegrowth.

2. distinct underlying modes of collaboration (collaborative vs.atomistic).

3. effects of the underlying social network as a recruitment factor insocial media groups.



Group affiliations vs. group-centered social network

G1 G2A.

SNG2B.



Flickr dynamics: metrics considered

I Group metrics considered as independent variables:

demographic metricsu0 number of group membersp0 number of photosms average membership spread

structural metricsk average directed degreec3 average clustering coefficientr reciprocity index

governance metricsadm number of group adminsmod number of group moderatorsµ moderation filterθ throttling index



Flickr dynamics: growth metrics

I Growth metrics considered as dependent variables:

Absolute size variation ∆u population variation (u1 − u0)∆p content variation (p1 − p0)

Population turnover u+ users who joined the groupu− users who left the group

Note: relative growth rates (normalized by initial size) were discardedas they did not display any statistically significant effects



Flickr dynamics: regression analysis

population content population turnovervariation variation joining leaving

parameter value (p) value (p) value (p) value (p)intercept -0.27 > .1 -2.20 *** -1.02 *** -4.91 ***

u0 0.87 *** - 0.77 *** 0.78 ***p0 - 0.94 *** 0.11 *** -0.03 ***

r 0.99 *** 2.09 *** - -0.19 **c3 -1.87 *** -1.73 *** -1.49 *** -1.27 ***k 0.10 * 0.18 *** - 0.23 ***

ms -0.57 *** -0.33 *** -0.43 *** 0.35 ***

µ - - 0.08 ** 0.07 ***mod 0.05 ** 0.09 *** 0.08 *** 0.02 ***θ - 0.10 *** - -

adm/u0 - - -0.06 *** -R2 0.65 0.75 0.68 0.82

(*, ** and *** mean a p-value smaller than 0.05, 0.01 and 0.001 respectively)



Demography-driven growth enhancers

Do larger groups grow faster?

Initial group size: population (left) and content (right)

1H283.054L

2H793.722L

3H1407.33L

4H2224.54L

5H3497.20L

6H5595.47L

7H9603.87L

8H20553.0L

9H120041.L

p0quantilesHquantilemeansL

10

100

1000

absolutevariation

1H283.054L

2H793.722L

3H1407.33L

4H2224.54L

5H3497.20L

6H5595.47L

7H9603.87L

8H20553.0L

9H120041.L

p0quantilesHquantilemeansL

10

100

1000

absolutevariation



Social network-driven growth regulators

Do cohesive groups grow slower?

clustering coefficient (left) and link reciprocity (right)

1H0.031L

2H0.073L

3H0.11L

4H0.14L

5H0.17L

6H0.21L

7H0.25L

8H0.32L

9H0.47L

c3quantilesHquantilemeansL

5

10

50

100

500

1000

absolutevariation

1H0.23L

2H0.30L

3H0.33L

4H0.36L

5H0.39L

6H0.42L

7H0.46L

8H0.50L

9H0.59L

rquantilesHquantilemeansL

5

10

50

100

500

1000

absolutevariation



Social network-driven growth enhancers

Do groups with more socially active members grow faster?

Average degree

1H0.46L

2H1.0L

3H1.6L

4H2.4L

5H3.3L

6H4.7L

7H7.1L

8H12.L

9H31.L

kquantilesHquantilemeansL

10

20

50

100

200

500

1000

2000

absolutevariation



Governance-driven growth regulators

Do curated groups grow differently from regular groups?

throttling (left), admin density (right)

1H0.43L

2H1.2L

3H2.0L

4H3.0L

5H4.0L

6H5.2L

7H15.L

Θ

quantilesHquantilemeansL

20

50

100

200

500

1000

absolutevariation

1H0.00036L

2H0.0010L

3H0.0019L

4H0.0030L

5H0.0043L

6H0.0060L

7H0.0082L

8H0.014L

9H0.12L

adm�u0quantilesHquantilemeansL

1

2

3

4

%growth



Flickr group dynamics: conclusions

I distinct drivers behind absolute growth and user turnover

I strong correlation of growth landscapes for content andpopulation (consistent with wikis)

I no significant effects on relative growth rates (size matters)

I key role of network structure in enhancing or regulatinggrowth

I negligible role of governance-driven factors

D. Taraborelli, C. Roth (2009)Affiliates or friends? Drivers of group dynamics in online social media (inpreparation).



Modelling viable online communities

Preliminary conclusions and research directions

What’s next:

(1) realistic characterizations of online community viability(2) challenges to a unified theory of online community

dynamics(3) empirically grounded hypotheses on micro-level behaviour




(1) Empirical challenges for realistic models of the viability ofonline communities.

I we focused so far on monotonic group population growth as a mainindicator of viability.

I compared to systems where the definition of viable states is morestraightforward (e.g. ecosystems), viability in social system is stronglygoal-dependent

I online communities with different goals may treat different dynamics asviable, e.g.

I maintain monotonic population growth

I keep the number of leaving members below a desired rate

I maximize member turnover

I maintain a manageable content/population ratio




(2) Challenges to a unified framework to model collaborativeonline communities

I dramatic differences in the effect of governance policies on differenttypes of collaborative systems.

I we proposed that this may be due to differences in the underlyingcollaborative mode

I atomistic in social media(users contribute to a common good by maintaining control over their contribution, e.g. photos

shared in a group pool)

I incremental in peer production systems(users contribute to a common good collaboratively authored by a collective of users where

maintaining control over individual contributions is not possible, e.g. edits to a wiki page or code

modifications to an open source project)




(3) Need of realistic hypotheses on micro-level behaviour

I Most existing models of group affiliation and social networkcoevolution in online communities focus on recruitmentdrivers such as:

I social influence

I social fringe recruitment

(Backstrom et al. 2006; De Choudury 2009; Hui et al. 2009)

I However this assumption can be challenged on empiricalgrounds (Zheleva et al. 2009)

I Need of realistic hypotheses of what drives joining andleaving behaviour.


Viability of Online Communities -...

Documents

Transcript of Viability of Online Communities -...