Coevolution of knowledge networks and
21st century cyberinfrastructure
Noshir Contractor
Professor, Departments of Speech Communication & Psychology
Co-Director, Age of Networks, Initiative, Center for Advanced Study
Director, Science of Networks in Communities National Center for Supercomputing Applications
University of Illinois at Urbana-Champaign
nosh@uiuc.edu
1.
Turn on power & set MODE with MODE button. You can confirm
the MODE you chose as the red indicator blinks.
2.
Lamp blinks when (someone with) a Lovegety for the opposite sex
set under the same MODE as yours comes near.
3.
FIND lamp blinks when (someone with) a Lovegety for the
opposite sex set under different mode from yours comes near.
May try the other MODES to “GET” tuned with (him/her) if you like.
Aphorisms about Networks

Social Networks:


Cognitive Social Networks:


Its not what you know, its who you know.
Its not who you know, its who they think you know.
Knowledge Networks:

Its not who you know, its what they think you
know.
Cognitive Knowledge Networks
Source: Newsweek,
December 2000
Amazon Purchase Network of Books on “Network Theory”
Amazon buyers Network of Top Selling Books on “Network Science”
Amazon buyers Network of Top Selling Books on “Network Society”
TECLab/SONIC Projects
on Enabling Networks

Networks to enable Cyberinfrastructure, NCSA/NSF

Emergency Response Networks, NSF-ITR

Tobacco Surveillance, Research & Evaluation Networks, NCI/NIH

Transnational Immigrant Networks, Rockefeller Foundation

Economic Justice Networks, Rockefeller Foundation

Communities of Practice Networks, Procter &Gamble

Food Safety Networks, UIUC Cross-Campus Initiative & John Deere

Global Supply Chain Infrastructure, Vodafone
Science and Engineering Cyberinfrastructures
Geosciences Cyberinfrastructures
SEEK: The Science
Environment for
Ecological Knowledge
Testbed Communities: Partners




Collaborative for Large-scale Engineering
Analysis Network for Environmental Research
(CLEANER): Barbara Minsker, UIUC
Tobacco Systems Integration Grid (Tobacco
SIG): Scott Leischow, NCI
Social Network Analysis CI (SNAC): Katy
Borner, Indiana U
Engaging People in Communities (EPIC):
Scott Lathrop, NCSA Education & Outreach
TECLab/SONIC Projects
on Enabling Networks

Networks to enable Cyberinfrastructure, NCSA/NSF

Emergency Response Networks, NSF-ITR

Tobacco Surveillance, Research & Evaluation Networks, NCI/NIH

Transnational Immigrant Networks, Rockefeller Foundation

Economic Justice Networks, Rockefeller Foundation

Communities of Practice Networks, Procter &Gamble

Food Safety Networks, UIUC Cross-Campus Initiative & John Deere

Global Supply Chain Infrastructure, Vodafone
{
ICT Support in
Emergency Management
Networks
Drawing Analogies
from Natural
Systems
Natural System: Honey Bees
ENTOMOLOGY: Learning from
natural robust societies.
Successful systems (evolution
time)
Ant - based models have
successfully been applied to
solve optimization [Dorigo, 1996;
Botee, 1999] and networking [Bonabeau,
2000] problems, among others.
Bees’ setting and objectives in foraging [Seeley, et al. 1991] resembles disaster
relief response scenario (collective decision-making).
Problem: Information Overload

Hundreds or Thousands of first
responders operate sharing
couple of voice channels (radio,
cell-phones) [Domel, 2001]
http://www.hollandsentinel.com/images/031503/Borculofire4.jpg

If technology provides a mean to enhance delivery and media of
information, we envision this problem would increase
Information Overload: Ants
Analogy (Ants’ alarm propagation)
Division of Labor; each ant “has”
a threshold for each stimulus
(pheromone).
When stimulus is greater than
threshold the ant will be on
“alarm” mode.
Centels ants detects a hazard and
release
“alarm”
pheromone
(volatile).
Each pheromone release will last
for a limited time; seconds or
minutes.
The heterogeneous response to
alarm pheromone avoids all ants
react immediately (good or bad?).
H
O
W
Idea:
Actors will propagate information
received only if the stimulus, i.e.,
“quality of information”, is greater than
his/her threshold for that type of
information.
Avoiding cascading effect; controlling
information overload.
Natural System: Honey Bees
Honey Bees (Apis melifera)
Foraging Model [Seeley, 1991]
At hive unloading
nectar from A
(HA)
p3
At hive unloading
nectar from B
(HB)
p1
p5
fxA
p7
fx B
1-f x A
1-f x B
Following
other dances
(F)
fd A(1-f x A)
Dancing for A
(DA )
fd B(1-fx B)
Dancing for B
(DB )
(1-f dA)(1-f x A)
(1-f dB)(1-f x B)
p4
The system evaluates
ALL the information,
though individuals
evaluate only partial
information
p2
p6
ff A
Foraging at nectar
source A
(A)
ff B
Foraging at nectar
source B
(B)
TECLab/SONIC Projects
on Enabling Networks

Networks to enable Cyberinfrastructure, NCSA/NSF

Emergency Response Networks, NSF-ITR

Tobacco Surveillance, Research & Evaluation Networks, NCI/NIH

Transnational Immigrant Networks, Rockefeller Foundation

Economic Justice Networks, Rockefeller Foundation

Communities of Practice Networks, Procter &Gamble

Food Safety Networks, UIUC Cross-Campus Initiative & John Deere

Global Supply Chain Infrastructure, Vodafone
INTERACTION NETWORKS
Non Human Agent to
Non Human Agent
Communication
Non Human Agent
(webbots, avatars, databases,
“push” technologies)
To Human Agent
Publishing to
knowledge repository
Retrieving from
knowledge repository
Human Agent to Human Agent
Communication
Source: Contractor, 2001
COGNITIVE KNOWLEDGE NETWORKS
Non Human Agent’s
Perception of Resources
in a Non Human Agent
Human Agent’s Perception of
Provision of Resources in a
Non Human Agent
Non Human Agent’s
Perception of what a Human
Agent knows
Human Agent’s Perception of
What Another Human Agent
Knows
Source: Contractor, 2001
Human A
Human B
Human C
Non
Human
Agent X
Non
Human
Agent Y
Human A
Human B
Human to Human
Interactions and
Perceptions
Human to Non
Human Interactions
and Perceptions
Non Human to
Human Interactions
and Perceptions
Non Human to Non
Human Interactions
and Perceptions
Human C
Non Human
Agent X
Non Human
Agent Y
WHY DO WE
CREATE,
MAINTAIN,
DISSOLVE, AND
RECONSTITUTE OUR
COMMUNICATION AND
KNOWLEDGE NETWORKS?
Monge, P. R. & Contractor, N. S. (2003). Theories of
Communication Networks. New York: Oxford University
Press.
Why do actors create, maintain,
dissolve, and reconstitute network
links?



Theories of self-interest
Theories of social and
resource exchange
Theories of mutual
interest and collective
action





Theories of contagion
Theories of balance
Theories of homophily
Theories of proximity
Theories of co-evolution
Sources:
Monge, P. R. & Contractor, N. S. (2003). Theories of Communication Networks. New York:
Oxford University Press.
Contractor, N. S., Wasserman, S. & Faust, K. (in press). Testing multi-theoretical
multilevel hypotheses about organizational networks: An analytic framework and
empirical example. Academy of Management Review.
ANALYZING & ENABLING NETWORKS IN
CYBERINFRASTRUCTURE
1. Extend theories to predict
the dynamics of a
cybercommunity
(MTML+Entomology+
Epidemiology + ?)
Generative mechanisms
2. Develop agent-based
computational models to assess
and evaluate alternative scenarios
for the long term dynamics of
the cybercommunity
(Blanche)
Model predictions of cybercommunity
Competence-based design of Cyberinfrastructure
4. Develop and introduce
“cyberinfrastructure”
networking tools to enable the
cybercommunity
(Adhoc/Sensor networks, IKNOW)
5. Statistical methods to
empirically validate the
dynamics of the
cybercommunity as predicted by
theories and models
(p*/ERGM and MCMC techniques)
Iterative Refinements to theories about
dynamics of cybercommunity
Multi-level hypotheses and
concepts to be measured
3. Collect longitudinal empirical
data from participants in
cybercommunity
(KAME/NAME)
Web-based surveys and real time
observation and computer-captured data from
cybercommunity activities
Co-evolution of knowledge networks
and 21st century organizational forms
NSF KDI Initiative 1999-04. PI: Noshir
Contractor, University of Illinois.
 Co-P.I.s: Monge, Fulk, Bar (USC), Levitt,
Kunz (Stanford), Carley (CMU), Wasserman
(Indiana), Hollingshead (Illinois).
 Three dozen industry partners (global, profit,
non-profit):

 Boeing,
3M, NASA, Fiat, U.S. Army, American
Bar Association, European Union Project Team,
Pew Internet Project, etc.
Public Goods / Transactive
Memory
–Allocation to the Intranet
–Retrieval from the Intranet
–Perceived Quality and
Transactive Memory


Perception of Other’s
Knowledge
Communication to
Allocate Information
Quantity of Contribution to
the Intranet
Communication to
Retrieve Information
Inertia Components
Social Exchange
–Collaboration
- Retrieval by coworkers on
other topics
–Co-authorship
–Communication
Proximity
-Work in the same location
Integrating exogenous and endogenous
processes based on multiple theories at
multiple levels leads to many possible
realizations of the network.
p* Framework



The observed network is one realization of the many
possible random realizations of the network.
Confirmatory Network Analysis: The questions of
interest in statistical modeling is whether the
observed network exhibits the theoretically
hypothesized structural tendencies.
The statistical estimates of p* parameters indicate
whether network realizations with the theoretically
hypothesized properties have significantly large
probabilities of being observed in the network data
collected.
Source: Contractor, N. S., Wasserman, S. & Faust, K. (in press). Testing
multi-theoretical multilevel hypotheses about organizational networks:
An analytic framework and empirical example. Academy of Management
Review.
Modeling p* Random Graph Distributions

For an observed network, which we consider to be a realization
x of a random array X, we assume the existence of a
dependence graph D for the random array X.

The edges of D are crucial here; consider the set of edges, and
determine if there are any complete subgraphs, or cliques found
in the dependence graph.

For a general dependence graph, a subset A of the set of
relational ties ND is complete if every pair of nodes in A (that is,
every pair of relational ties) is linked by an edge of D. A subset
comprising a single node is also regarded as complete.

These cliques specify which subsets of relational ties are all pair
wise, conditionally dependent on each other.
Source: Carrington, P., Scott, J., & Wasserman, S. (Eds.). (2005).
Models and Methods in Social Network Analysis. New York: Cambridge
University Press.
Using Dependence Graphs to Model
p* Random Graph Distributions

The Hammersley-Clifford theorem (Besag, 1974) provides the important link between the
dependence graph and the structure of the model that encapsulates its dependence
assumptions.

The theorem establishes that the probability model for the random multigraph, X, depends
on the complete subgraphs of the dependence graph, D.

A complete subgraph, or clique, is a subset of nodes in the dependence graph every pair of
which is linked by an edge. A subset consisting of a single node is also regarded as
complete.

Each complete subgraph corresponds to a configuration of possible ties in the network.

There is a model parameter corresponding to each complete subgraph in the dependence
structure (and so to each corresponding configuration of possible ties).

The parameter for a particular configuration reflects the effect of observing that configuration
on the likelihood of the network.
Using Dependence Graphs to Model
p* Random Graph Distributions

The random graph model is of the following exponential form:

Pr(X = x) = p*(x) = κ-1 exp∑ANDλAzA(x)}
where:
x is a realization of the random graph, X;
κ = ∑x exp∑ANDλAzA(x)} is a normalizing quantity; the summation is over all
subsets A of nodes of D;
z A(x) is the empirically observed network statistic in x corresponding to the
subgraph A of D and is given by z A(x) = Π Xij∈A xij;
λA is the parameter corresponding to the subgraph A of D; and λA = 0 whenever the
subgraph induced by the nodes in A is not a complete subgraph of D.
Interpreting Parameters in the Model
p* Random Graph Distributions

The random graph model is of the following exponential form:
Pr(X = x) = p*(x) = κ-1 exp∑ANDλAzA(x)}

The quantities zA(x) are calculated from the observed network
and correspond to the hypothesized structural tendencies
expressed in the dependence graph.

λA are parameters corresponding to the cliques A of D. These
parameters express the importance of the associated structural
tendency for the probability of the graph.
Motivation for Information Retrieval in Knowledge Networks
1. Social Communication
0.144
2. Perception of Knowledge
& Communication to Allocate
0.995
3. Perception of Knowledge & Provision
0.972
4. Perception of Knowledge, Social Exchange,
& Social Communication
0.851
5. Perception of Knowledge, Proximity,
& Social Communication
0.882
3D Vision for SONIC



Discovery tools to effectively and efficiently foster
network links from people to other people,
knowledge, and artifacts (data sets/streams, analytic
tools, visualization tools, documents, etc.) within the
cybercommunities.
Diagnostic tools to assess the “health” of knowledge
networks within cybercommunities: scanning,
absorptive capacity, diffusion, robustness,
vulnerability.
Design or re-wire networks using social and
organizational incentives as well as computationally
advanced and intensive network referral systems to
enhance evolving and mature communities.
IKNOW Demo
Summary



21st century cyberinfrastructure, like the Lovegety,
necessitates studying the emergence – creation,
maintenance, dissolution, and reconstitution – of
networks.
Research on emergence of networks requires an
analytic approach that empirically tests the
simultaneous influence of multi-theoretical
explanations at multiple levels.
p* /ERGM confirmatory network analytic methods
have proven useful in simultaneously testing
hypotheses using this framework.
Contact information
nosh@uiuc.edu
www.uiuc.edu/ph/www/nosh