ArticlePDF Available

Smartocracy: Social Networks for Collective Decision Making

Authors:

Abstract and Figures

Smartocracy is a social software system for collective decision making. The system is composed of a social network that links individuals to those they trust to make good decisions and a decision network that links individuals to their voted-on solutions. Such networks allow a variety of algorithms to convert the link choices made by individual participants into specific decision outcomes. Simply interpreting the linkages differently (e.g. ignoring trust links, or using them to weight an individual?s vote) provides a variety of outcomes fit for different decision making scenarios. This paper will discuss the Smartocracy network data structures, the suite of collective decision making algorithms currently supported, and the results of two collective decisions regarding the design of the system.
Content may be subject to copyright.
Smartocracy:
Social Networks for Collective Decision Making
Marko A. Rodriguez1, Daniel J. Steinbock2, Jennifer H. Watkins1, Carlos Gershenson3,
Johan Bollen1, Victor Grey4, Brad deGraf5
1Los Alamos National Laboratory, Los Alamos, New Mexico 87545
2Stanford University, Stanford, California 94305
3Vrije Universiteit Brussel, Brussel, Belgium 1160
42idi Corporation, Concord, California 94524
5Media Venture Collective, Mill Valley, California 94941
Email: marko@lanl.gov
Abstract
Smartocracy is a social software system for collec-
tive decision making. The system is composed of a so-
cial network that links individuals to those they trust to
make good decisions and a decision network that links
individuals to their voted-on solutions. Such networks
allow a variety of algorithms to convert the link choices
made by individual participants into specific decision
outcomes. Simply interpreting the linkages differently
(e.g. ignoring trust links, or using them to weight an
individual’s vote) provides a variety of outcomes fit for
different decision making scenarios. This paper will
discuss the Smartocracy network data structures, the
suite of collective decision making algorithms currently
supported, and the results of two collective decisions
regarding the design of the system.
1 Introduction
The recent explosion of so-called ‘social software’
on the Internet has been characterized by democratic
approaches to content generation [11]. The Wiki is an
exemplar of this approach, where all users have equal
power to add or modify the content of any hypertext
[5]. Wikis are social software that support democratic
collaborative authorship. Systems exist to support
generative collaboration in many fields including,
but not limited to, journalism1, scholarly citation2,
1Digg available at: http://digg.com
2CiteULike available at: http://citeulike.org
photography3, and hypertext bookmarking4.
While these systems vary in the forms of col-
laboration they support, they have in common an
egalitarian social structure and they all aggregate user
contributions into shared representations of collective
belief. For instance, in addition to contributing news
stories, users of the Digg web service vote for stories
they think highly of and can view the most popular
stories for different subjects. Del.icio.us users save
and categorize bookmarks for websites they like into
a common pool and can easily view the most popular
sites for any category. Similar patterns of contribution
and aggregation occur with varying prominence
throughout the social software sphere. Despite the
proliferation of such systems with traditional methods
of aggregation, there has yet to emerge a generalized
software model for the intelligent aggregation of
individual contributions beyond mere vote-counting.
If such a model did exist, it could systematically
improve the state of the art in social software design
and promote the innovation of systems geared more di-
rectly toward the aggregation of individual knowledge
into collective knowledge, i.e. software supported col-
laborative problem solving and decision making [2, 12].
Smartocracy5is a web-based social software sys-
tem for collective problem solving/decision making.
Smartocracy uses a problem-solution model where in-
dividuals pose problems (i.e. issues, questions) to the
3Flickr available at: http://flickr.com
4Del.icio.us available at: http://del.icio.us
5Smartocracy available at: http://www.smartocracy.net
Proceedings of the 40th Hawaii International Conference on System Sciences - 2007
1
U.S. Government Work Not Protected by U.S. Copyright
Smartocracy community and propose potential solu-
tions (i.e. options, answers). The proposed solutions
are voted on and the aggregate preferences of all in-
dividual users yield the collective decision. To further
facilitate vote-based decision making, a trust-based so-
cial network is used to represent the relations among
users and, in some cases, to support the automated del-
egation of decision making power along paths of trust.
This paper discusses trust-based decision making the-
ory, the Smartocracy system implementation, and con-
cludes with the results of the Smartocracy community’s
aggregate system development decisions as of March
2006.
2 Collective Decision Making
Collective decision making, in the context of Smar-
tocracy, involves individuals generating problems,
providing potential solutions, voting for solutions,
and the software aggregating individual votes and
ultimately deriving a final collective decision. Figure 1
provides an outline of Smartocracy’s collective decision
making process.
Collective
Solution Ranking
Individual
Solution Ranking
Final
Solution Selection
Generate Potential
Solutions
problem
INPUT
solution
OUTPUT
(optional)
Figure 1. Smatocracy’s collective decision
making stages
Given a particular problem, any individual in Smar-
tocracy can propose a potential solution. Individuals
can also vote for a proposed solution. By voting for
particular solutions, individuals are explicitly provid-
ing their individual ranking of the problem’s solution
set. The suite of Smartocracy collective solution rank-
ing algorithms aggregates all the individual votes into
a collective solution ranking. Collective ranking algo-
rithms currently supported by Smartocracy are direct
democracy [6], dynamically distributed democracy [8],
and proxy vote [9]. The result of the collective solution
ranking algorithm serves as input to a solution selec-
tion function. For nominal-, or categorical-, based solu-
tion sets, the highest ranked solution is considered the
collective decision. For numeric-, or gradient-, based
solution sets, a weighted average selection function is
calculated based on their respective rank. The result
returned by the selection function is the collectively
derived solution to the problem. It is important to
note that the solution selection function is an optional
step that is used only if a single definitive solution is
required. In some cases, the solution to the problem
may be the collective ranking itself.
3TheNetworks
Social networks are used to represent the rela-
tionships between individuals of a population. Most
commercial social network systems (e.g. MySpace6
and LinkedIn7) denote static relational ties between
users, such as ‘is a friend’ or ‘is a colleague’. For
collective decision making, friendship is not neces-
sarily the desired semantic relationship. In collective
decision making, stating that an individual is a
friend does not identify them as a good decision
maker [14]. Therefore, a collective decision making
social network must make explicit the notion of
decision making trust. For instance, if individual A
provides individual Bwith a trust edge, then Ais
stating that he or she trusts Bto make a good decision.
There exist other artifacts in the system besides
human individuals. These artifacts include problems
(i.e. current issues being addressed by the collective)
and solutions (i.e. potential options for a particular
problem). Therefore, the graph data structure under-
lying the Smartocracy system is a graph connecting
individuals according to their relative trust of one
another and connecting individuals to particular
solutions via edges that denote a vote. The network
data structure describing this system can be formally
denoted as G={N,W}where Nis a collection of
human (H), problem (P), and solution (S)nodes
and Wis the set of weighted semantic relationships
between these entities. Note that N=HPS,
HPS=,andwλ
ni,njstates that there exists a
directed, weighted relationship of semantic λbetween
niand nj. The only semantic relationships currently
supported by Smartocracy are λ={trusts,votedFor}.
The connection between the trust-based social
network and the vote-based decision network is rep-
resented in Figure 2 where the solid edges represent
the weighted trust relations amongst individuals
and the dashed edges represent the weighted votes
that individuals provide for particular solutions to
problems. The following two sections will further
discuss each network’s constructs.
6MySpace available at: http://www.myspace.com
7LinkedIn available at: http://www.linkedin.com
Proceedings of the 40th Hawaii International Conference on System Sciences - 2007
2
trust-based social network
for domain 1
problem 1
problem n
...
...
...
domain 1
a potential
solution
Figure 2. The connection between the social
and decision networks
3.1 Trust-Based Social Network
In the Smartocracy trust-based social network, an
edge going from individual hiHto hjHstates
that hibelieves that individual hjis ‘good’ at decision
making where i=jand HN.Inthismodel,hjmay
not trust the decision making behavior of hi.There-
fore, the trust-based edges between individuals are di-
rectional. Furthermore, individual himay weight the
relative trustworthiness of individual hjand, in such
cases, an edge can be represented by the conditional
probability,
wtrusts
hi,hj=P(hjis good |hi’s knowledge of hj).
This relationship states that hitrusts, according to
some probability, that hjwill make a good decision
given hi’s previous understanding of hj. What this
edge model lacks is a representation of the domain
for which trust is given. Because humans are multi-
dimensional entities composed of various skills and be-
liefs, individual himay trust the decision making be-
havior of hjin one domain but not another. Therefore,
it may be more appropriate, given a decision making
scenario, to state,
wtrusts
hi,hj=
P(hjis good in domain dl
|hi’s knowledge of hjin domain dl).
There are multiple ways to represent domains in a
social network context [9]. Currently, due to the fo-
cused use of Smartocracy, there exists only a single so-
cial network and therefore, no explicit representation
of domains. As will be demonstrated in the results
section, domain specific representation is a collectively
desired feature for future implementation.
3.2 Vote-Based Decision Network
In Smartocracy, any individual that votes for a po-
tential solution is creating a directed edge from them-
selves in Hto a solution in S. The semantic of this
relationship is λ= votedFor and can be represented by
the conditional probability,
wvotedFor
hi,pj(sm)=
P(smis a good solution for problem pj
|hi’s knowledge of pj).
The above vote edge states that human hibelieves
that, according to some probability, smis a good
solution for problem pj.
The weighted ranking that an individual provides for
the solution set to a particular problem is their subjec-
tive evaluation of the relative optimality of the solu-
tions for the problem. These are called the individual
solution rankings. To move from an individual solution
ranking to a collective solution ranking, an aggregation
algorithm is required. The next section will discuss the
algorithmic framework used to calculate collective de-
cisions within the trust-based social network and the
vote-based decision network.
4 Algorithms for Collective Solution
Ranking
Given a trust-based social network and a vote-based
decision network, Smartocracy supports a family of
algorithms for aggregating individual votes into a
collective solution ranking. The currently imple-
mented algorithms are direct democracy, dynamically
distributed democracy, and proxy vote. All of these
algorithms are implemented under the parameterized
particle swarm framework described in [7, 9] and are
similar to the idea of constrained spreading activation
[1].
In a particle swarm, a particle is considered an
‘atom’ of decision making influence. To calculate a col-
lective solution ranking for a particular problem, parti-
cles begin their journey at human nodes and make their
way, in a stochastic manner, to the solution nodes of
Proceedings of the 40th Hawaii International Conference on System Sciences - 2007
3
that problem via the trust-based social network and
vote-based decision network. Because particles are
discrete, indivisible entities, the diffusion of particles
through a network requires a sufficient initial distribu-
tion to expose the underlying network topology. The
more particles initially supplied to the network, the
more accurate the collective ranking [7]. At the end of
the particle propagation algorithm, when all particles
have either been destroyed or have reached a solution,
the distribution of particles over the solution set of the
problem determines the collective’s solution ranking.
This section will describe the specifics of each of the
three collective solution ranking algorithms currently
supported by Smartocracy.
4.1 Direct Democracy
Direct democracy embodies the idea of ‘one-
person/one-vote’ [6]. In direct democracy, the
trust-based social network is not used to calculate
the collective decision. If the individual does not
vote, then the individual does not participate. If an
individual does not participate, then he or she does not
influence the collective solution ranking. To implement
this algorithm within the particle swarm framework,
each individual is supplied with 100 particles. A
particle can only traverse a votedFor edge, wvotedFor
hi,pj(sm).
The probability of traversing one votedFor edge over
another is dependent upon the edge weight assigned
by the voter. Strongly weighted edges have a higher
probability of being traversed by the particle. If the
individual has not voted, then the particle destroys
itself. After one step, all particles are either destroyed
or are at a particular solution node to problem pj.
The distribution of particles over the solution nodes
represents the direct democracy collective solution
ranking.
Figure 3 provides an example of the possible paths
of a direct democracy swarm. Humans h1,h2,and
h3are each provided 100 particles. After the first
step, because particle diffusion is a stochastic process
biased by edge weights, s1will have accumulated
approximately 50 particles, s260 particles, and s3
90 particles. Note that the particles given to h2are
destroyed because h2has not voted on a solution.
The normalized distribution over the solution set is
s1=0.25, s2=0.30, and s3=0.45. This normalized
distribution is the direct democracy collective solution
ranking. If the solution set is a nominal solution set,
then s3is the collectively derived solution according
to a highest rank solution function.
12
3solutions
to problem
human
collective
123
0.50
0.50
0.10 0.90
Figure 3. Direct democracy particle paths
4.2 Dynamically Distributed Democracy
Dynamically distributed democracy, or DDD,
was developed to handle fluctuating levels of par-
ticipation. In such cases, ad hoc representative
structures are created to ensure that every individual
can influence the collective solution ranking even if
only through a proxy representative. This algorithm
has been shown to be an accurate way to model the
collective’s perspective as voter participation wanes [8].
In DDD, a particle, if it is unable to take a vote
edge to a particular solution, uses the trust-based
social network to move to a proxy representative. If
that representative has voted, then the particle moves
to one of the representative’s chosen solutions. If the
representative has not voted, the particle traverses
a trust edge to move to yet another representative.
This iterative process continues until a solution to the
problem is found.
In Figure 4, human h2has not voted on a particular
solution. In DDD, h2’s decision making influence
(in the form of particles) is delegated to human h1
because h2trusts h1,wtrusts
h2,h1. Therefore, h1has 200
particles to provide to his or her chosen solutions.
After two steps, solution s1will have 100 particles, s2
will have 110 particles, and s3will have 90 particles.
The normalized distribution over the solution set is
s1=0.33, s2=0.36, and s3=0.30. If the selection
function selects the highest ranked solution, then
solution s2would be the collectively derived solution.
Notice that because h3provided a vote, the particles
initially provided to h3do not traverse the social net-
work. That is, wtrusts
h3,h2isnotusedintheDDDcompu-
tation. Furthermore, in the case that every individual
votes for a particular solution, the DDD algorithm will
provide a collective solution ranking that is equivalent
to the direct democracy algorithm’s ranking. DDD is
Proceedings of the 40th Hawaii International Conference on System Sciences - 2007
4
12
3solutions
to problem
human
collective
123
0.50
0.50
0.10 0.90
Figure 4. Dynamically distributed democracy
particle paths
useful in problem domains where it is desirable for ev-
ery individual to have an equal say (i.e. influence) in
the collective’s decision, but it is not feasible for every
individual to actively participate via voting. In such
cases of reduced participation, ad hoc representative
structures emerge to simulate full participation.
4.3 Proxy Vote
Proxy vote is an extension of DDD where the initial
distribution of particles is biased by the trust-based
in-degree of an individual. That is, the more a
particular individual is trusted, the more particles
that individual initially receives. This algorithm is
used for expert-based problem domains where an
equal say for every individual is not desired. For proxy
vote, it is assumed that the in-degree of the indi-
vidual is a representation of his or her level of expertise.
In Figure 5, human h1is supplied with 200 particles,
h2200 particles, and h3100 particles. The number of
particles at h1and h2is double that of h3because both
h1and h2have one incoming trust-based edge and each
incoming edge provides an extra 100 particles to an
individual. As in the DDD example, h2disseminates
his or her received particles to h1because h2trusts only
h1and h2has not voted on a solution to the problem.
Therefore, h1has 400 particles to distribute to his or
her chosen solutions. After the second step, solution
s1will have 200 particles, s2210 particles, and s3will
have 90 particles. The normalized distribution over
the solution set is s1=0.40, s2=0.42, and s3=0.18.
Given a highest rank selection function, s2would be
the collective decision.
5 Smartocracy Beta Results
Smartocracy was released for beta testing in
February of 2006. New users join the system via an
12
3solutions
to problem
human
collective
123
0.50
0.50
0.10 0.90
Figure 5. Proxy vote particle paths
invitation from a currently existing user. An invitation
of someone new automatically creates a directed edge
from the inviter to the invitee. Throughout the life
of the system, individuals can reassign their outgoing
trust-based edges. The interface to view one’s incom-
ing and edit one’s outgoing trust edges is shown in
Figure 6.
Figure 6. An individual’s in and out trust-
based edges
Figure 7 provides a visualization of the 276 partic-
ipants in the Smartocracy trust-based social network
as of March 7, 2006.
The current in-degree edge distribution is presented
Proceedings of the 40th Hawaii International Conference on System Sciences - 2007
5
264
2
82
211
222
265
266
267
268
269
270
35
68
32
36
81
33
72
147
65
162
96
160
70
5
43 140
210
220
221
223
224
225
226
263
52
57
208
55
153
271
184
185
186
187
188
189
190
191
192
194
193
242
230
227
229
237
272
182
207
214
273
212
213
216
217
218
219
87
197
198
88
97
203
204
205
3
195
196
1
44
40
49
41
42
46
89
132
77
114
98
199
50
51
144
145
146
148
118
119
120
121
122
123
258
259
106
107
108
109
110
111
60
61
62
63
64
232
235
236
245
246
247
248
154
151
155
156
157 158
159
80
84
141
142
143 150
116
115
117
172
173
174
175
176
177
178
179
180
181
112
71
201
202
240
241
249
125
126
124
86
133
134
135
136
137
138
139
206
90
260
261
262
274
275
276
168
169
170
171
66
250
257
251
252
253
254
255
256
0
67
69
74 75
76
83
94
113
161
200
228
244
85
91
92
93
163
164
165
99 100
101
102
103
104
127
128
129
130
131
Figure 7. Visualization of the Smartocracy
network as of March 7, 2006
on a log/log plot in Figure 8. Like most real world
networks, the Smartocracy social network’s edge
distribution follows a power-law. For the function
P(in-degree) = a×in-degreeγ,γ2.4. This function
computes the probability that any given node will
have a particular in-degree. As can be seen in Figure
8, there exist many individuals with few incoming
trust edges and few individuals with many incoming
edges. As depicted in Figure 8, more than 75% of all
individuals in the network have 1 incoming edge while
only 0.4% of the individuals have 8 incoming edges.
12345678
0.005 0.020 0.100 0.500
trust−based in−degree
pro
b
a
bilit
y
Figure 8. Smartocracy social network in-
degree edge distribution
The remainder of this section will provide some pre-
liminary results of two initial problems posed to the
Smartocracy collective. The two problems, in the form
of questions, are:
1. What should the name of the system be?
2. What features should we add to the system?
These problems are self-reflective, as they concern
the system itself. The goal of the beta test period is
to develop a system that is palatable to the collective.
For this reason, it is thought best to allow the users to
provide development solutions and thus regulate the
system’s evolution via the collective. It is important
to note that the results of the various collective solu-
tion ranking algorithms differ. We are not attempting
to state whether one algorithm is more optimal than
another, only to demonstrate that different metrics for
collective decision making exist and that it is impor-
tant to study in which contexts which algorithms are
most appropriate.
5.1 System Name
The system name problem was posed by the
developers in order to determine the most appropriate
name for the service. Any individual could provide
a potential name solution as well as vote on which
name they preferred. Table 1 provides the 13 potential
namesthatwereprovidedbythegroup.
potential solution
1DecisionNet
2DDD
3Dis Dis Sys
4Cell of the Elite
5Smart Mob Rule
6Smartocracy
7Holocracy
8netocracy
9Wisism
10 Stupocracy
11 Cheerocracy
12 antpile
13 Decision Network
Table 1. System name solutions
A visualization of the vote interface of the system
is provided in Figure 9. Between the time the initial
decision making data was analyzed and the screenshot
was taken, a new system name, Dynamocracy, was
proposed. Note that the new solution expresses the
dynamic nature captured in the difference between
Table1andFigure9.
Figure 10 provides the ranking of the 13 potential
solutions according to the direct democracy algorithm.
Of the 276 participants in the system, 95 voted.
Therefore, only 34% of the group contributed to the
Proceedings of the 40th Hawaii International Conference on System Sciences - 2007
6
Figure 9. Creating and voting on solutions
screenshot
direct democracy collective solution ranking. With a
highest rank selection function, the name Smartocracy,
is the final collective decision.
12345678910111213
0.0 0.1 0.2 0.3 0.4 0.5 0.6
Figure 10. Direct democracy system name
ranking
Figure 11 provides the ranking of the 13 potential
solutions according to the dynamically distributed
democracy algorithm. In DDD, the 181 non-
participants delegated their decision making influence
(particles) to their trusted proxy representatives.
Given a highest rank selection function, Smartocracy
is again the collective decision.
Finally, Figure 12 provides the ranking of the
1 2 3 4 5 6 7 8 9 10 11 12 13
0.0 0.1 0.2 0.3 0.4 0.5 0.6
Figure 11. Dynamically distributed democ-
racy system name ranking
13 potential solutions according to the proxy vote
algorithm. Smartocracy is once again the highest
ranked solution.
1 2 3 4 5 6 7 8 9 10 11 12 13
0.0 0.1 0.2 0.3 0.4 0.5 0.6
Figure 12. Proxy vote system name ranking
Table 2 provides the Spearman ρcorrelations be-
tween the various algorithms’ collective solution rank
distributions where dd is direct democracy, ddd is dy-
namically distributed democracy, and pv is proxy vote.
The strong correlations, with p-value <2.216,be-
tween the algorithms demonstrates the definitiveness
of the collective’s desired name choice. For this reason,
Smartocracy is the current name of the system.
dd ddd pv
dd 1.0 0.93 0.91
ddd 0.93 1.0 0.97
pv 0.91 0.97 1.0
Table 2. System name Spearman ρrank cor-
relations
Proceedings of the 40th Hawaii International Conference on System Sciences - 2007
7
5.2 Future Features
In line with today’s widely-used strategies for social
software development, Smartocracy was originally
implementedinabare-bonesfashionwiththedesire
that the group steer its future development. With this
idea in mind, the problem “What features should we
add to the system?” was posed to the group. Table 3
provides the list of all proposed solutions generated by
members of the group.
potential solution
1social network visualization
2know if my proxies have taken action
3domain specific proxies
4Golightly Community
5dialog with other participants
6geo-based network viz
Table 3. Future features solutions
Of the 276 participants in the system, 98 voted
for a future feature. Figure 13 depicts the direct
democracy collective solution ranking over the 6
proposed solutions. According to direct democracy
with a highest rank selection function, the collective
desires to have a feature that allows individuals to
know if their representatives, or proxies, have voted
for a solution to a problem.
123456
0.00 0.10 0.20 0.30
Figure 13. Direct democracy collective solu-
tion ranking for features
In Figure 14, the DDD collective solution ranking
provides a different perspective of the collective’s
desired future feature. Smartocracy was originally
developed with a single trust-based social network.
Therefore, decision making influence is delegated in
the system irrespective of the domain of the problem.
It became apparent to many users that they may
trust someone in one domain, but not another. The
DDD/highest rank decision shows that the collective
is more interested in having domain specific proxies,
or domain specific representatives. That is, decision
making influence is delegated to representatives de-
pending on the domain of the problem and the domain
of the trust-based representation.
123456
0.00 0.10 0.20 0.30
Figure 14. Dynamically distributed democ-
racy collective solution ranking for features
Finally, the proxy vote solution ranking is perfectly
correlated with the DDD solution ranking and there-
fore, proxy vote shows that the collective is interested
in having domain specific proxies. The proxy vote
collective solution ranking is provided in Figure 15.
123456
0.00 0.10 0.20 0.30
Figure 15. Proxy vote collective solution
ranking for features
The Spearman ρcorrelations between the various
algorithm’s collective solution rankings are provided in
Table 4 w h e r e p-value <2.216. Given the perfect cor-
relation between DDD and proxy vote, the developers
of Smartocracy are currently working towards imple-
menting domains into the system as articulated in [9].
Proceedings of the 40th Hawaii International Conference on System Sciences - 2007
8
dd ddd pv
dd 1.0 0.88 0.88
ddd 0.88 1.0 1.0
pv 0.88 1.0 1.0
Table 4. Future features algorithm Spearman
ρcorrelations
6 Future Directions
In spirit with the previously presented collective
decision making processes, future development of the
system will be guided by the requirements generated
by the collective. As shown in the future features
section, a ma jor push in the development of this
system will be the inclusion of domain specific repre-
sentation. In [9], a model for adding domain specific
representation to social networks is provided. Depend-
ing on the requirements of the Smartocracy group,
one of the proposed models will be implemented.
Furthermore, the current platform can be extended
to support collaborative discourse as presented in
[13, 12]. Collaborative discourse promotes interaction
amongst the participants by means of argumentation.
Finally, in terms of future research in the area of
collective decision making, we intend to explore vari-
ous collective ranking algorithms for different classes
of problems to understand which algorithms are best
suited for particular decision making situations. Given
a problem set with known optimal solutions, which al-
gorithms best aggregate the perspectives of the individ-
uals to yield the optimal solution? Such a study would
expand the use of Smartocracy into the domain of col-
lective intelligence research and would bring Smartoc-
racy to a level that is a generalization of domain specific
human collective intelligence systems like [3, 4, 12, 10].
7 Conclusion
Social systems, in order to make the ‘right’ deci-
sions, require sensors capable of delivering relevant
information. Polls have been used to collect infor-
mation relevant for societal-scale decision making.
The drawbacks of polls are well known: either they
require public participation from a large sector of the
population, or they might be biased and therefore, not
representative of the collective opinion. By exploiting
social network structures, systems such as the one
presented here can overcome some of these drawbacks.
With many collective solution ranking algorithms,
other than just direct democracy, it is possible to
learn how these metrics effect the collective solution
outcome. In turn, over time it may be possible to
understand in which context which algorithm is best.
Smartocracy could be used to make social decisions
within communities of any type, from frivolous to sci-
entific, from political to educational: scientific peer
review, project funding, school admissions, popular-
ity contests, artistic awards, ethical decisions, judicial
sentences. These are only some of the potential niches
where the research presented in this paper could be
exploited.
8 Acknowledgments
The Smartocracy project was organized by Brad de-
Graf of the Media Venture Collective8whose vision is
to revolutionize the media to better serve public inter-
est through grassroots collective action. This project is
funded by grants from the Threshold Foundation and
the Wallace Global Fund.
References
[1] F. Crestani and P. L. Lee. Searching the web by con-
strained spreading activation. Information Processing
and Management, 36(4):585–605, 2000.
[2] F. Heylighen. Collective intelligence and its implemen-
tation on the web: Algorithms to develop a collective
mental map. Computational & Mathematical Organi-
zation Theory, 5(3):253–280, 1999.
[3] N. Johnson, S. Rasmussen, C. Joslyn, L. Rocha,
S. Smith, and M. Kantor. Symbiotic intelligence: self-
organizing knowledge on distributed networks driven
by human interaction. In 6th International Conference
on Artificial Life, 1998.
[4] C. A. Kaplan. Collective intelligence: A new ap-
proach to stock price forecasting. In Proceedings of
the 2001 IEEE Systems, Man, and Cybernetics Con-
ference, New York, USA, 2001.
[5] B. Leuf and W. Cunningham. The Wiki way: quick
collaboration on the Web. Addison-Wesley Longman
Publishing Co., 2001.
[6] A. Lijphart. Democracies: Patterns of Majoritar-
ian and Consensus Government in Twenty-One Coun-
tries. Yale University Press, 1984.
[7] M. A. Rodriguez and J. Bollen. Simulating network
influence algorithms using particle-swarms: Pagerank
and pagerank-priors. [submitted], 2005.
[8] M. A. Rodriguez and D. Steinbock. A social network
for societal-scale decision-making systems. In NAAC-
SOS ’04: Proceedingss of the North American Asso-
ciation for Computational Social and Organizational
Science Conference, Pittsburgh, PA, USA, 2004.
8Media Venture Collective available at
http://www.mediaventure.org
Proceedings of the 40th Hawaii International Conference on System Sciences - 2007
9
[9] M. A. Rodriguez and D. J. Steinbock. The anatomy of
a large scale collective decision making system. Techni-
cal report, Los Alamos National Laboratory Technical
Report [LA-UR-06-2139], March 2006.
[10] D. Steinbock, C. Kaplan, M. A. Rodriguez, J. Diaz,
N. Der, and S. Garcia. Collective intelligence quani-
tifed for computer-mediated group problem solving.
Technical report, University of California at Santa
Cruz, 2004.
[11] M. Tepper. The rise of social software. netWorker,
7(3):18–23, 2003.
[12] M. Turoff, S. Hiltz, H. Cho, Z. Li, and Y. Wang. So-
cial decision support systems (SDSS). In 35th Annual
Hawaii International Conference on System Sciences
(HICSS‘02), volume 1, page 11, Hawaii, USA, 2002.
[13] M. Turoff, S. R. Hiltz, M. Bieber, J. Fejermestad, and
A. Rana. Collaborative discourse structures in com-
puter mediated group communications. In 32nd An-
nual Hawaii International Conference on System Sci-
ences (HICSS‘02), volume 1, Hawaii, USA, 1999.
[14] C.-N. Ziegler and J. Golbeck. Investigating correla-
tions of trust and interest similarity - do birds of a
feather really flock together? Decision Support Sys-
tems, [in press], 2006.
Proceedings of the 40th Hawaii International Conference on System Sciences - 2007
10
... Credulix shares similarities with reputation systems [46,61,63], in creating profiles for users (UCRs in Credulix) and in assuming that the future behavior of users will be similar to their past behavior. In our approach, however, users are not rated directly by other users. ...
Preprint
Today's social media platforms enable to spread both authentic and fake news very quickly. Some approaches have been proposed to automatically detect such "fake" news based on their content, but it is difficult to agree on universal criteria of authenticity (which can be bypassed by adversaries once known). Besides, it is obviously impossible to have each news item checked by a human. In this paper, we a mechanism to limit the spread of fake news which is not based on content. It can be implemented as a plugin on a social media platform. The principle is as follows: a team of fact-checkers reviews a small number of news items (the most popular ones), which enables to have an estimation of each user's inclination to share fake news items. Then, using a Bayesian approach, we estimate the trustworthiness of future news items, and treat accordingly those of them that pass a certain "untrustworthiness" threshold. We then evaluate the effectiveness and overhead of this technique on a large Twitter graph. We show that having a few thousands users exposed to one given news item enables to reach a very precise estimation of its reliability. We thus identify more than 99% of fake news items with no false positives. The performance impact is very small: the induced overhead on the 90th percentile latency is less than 3%, and less than 8% on the throughput of user operations.
... The popularity of social networking platforms has permanently changed the structure and working of traditional decision support systems. The use of social media is considered to have direct financial benefits for organizations [47], it offers a platform for transparent discussions leading to faster decision cycles and competition among entrepreneurs [45]. The decisionmaking procedure enabled by social media technology varies from traditional decision support literature. ...
Article
Full-text available
Release of a theatrical movie trailer is a major marketing practice and a considerable cost is associated with it. Evaluating the effectiveness of a theatrical movie trailer before its release could substantially contribute towards enriching its contents and economic value. The relationship between the emotional responses generated in response to a movie trailer cannot be effectively measured using traditional methods such as surveys and interviews. This paper proposes a framework to measure the effectiveness of movie trailers by measuring emotive response of viewers. A case study was conducted to study the impact of a movie trailer release on stock value of movie using virtual stock markets. Further, the case study investigated the impact of emotionally intense movie trailer over its stock price. Based on emotive content of trailers, few of the movie stocks experienced a surge of two hundred and 50 % while others experienced a marginal rise of five to 10 % only. The observed results indicated a direct relation between release of movie trailer, its emotive content and abnormal positive returns of a movie stock.
... The proposed model was applied in collective decisionmaking processes [35] to evaluate its operability in practical application systems. NetLogo [36], an integrated agent-based programmable modeling environment, is widely used to simulate applications in interactive systems. ...
Article
Full-text available
While social computing related research has focused mostly on how to provide users with more precise and direct information, or on recommending new search methods to find requested information rapidly, the authors believe that network users themselves could be viewed as an important social resource. This study concentrates on analyzing potential and dynamic user correlations, based on topic-aware similarity and behavioral influence, which may help us to discover communities in social networking sites. The dynamically socialized user networking (DSUN) model is extended and refined to represent implicit and explicit user relationships in terms of topic-aware features and social behaviors. A set of measures is defined to describe and quantify interuser correlations, relating to social behaviors. Three types of ties are proposed to describe and discover communities according to influence-based user relationships. Results of the experiment with Twitter data are used to show the discovery of three types of communities, based on the presented model. Comparison with six different schemes and two existing methods demonstrates that the proposed method is effective in discovering influence-based communities. Finally, the scenario-based simulation of collective decision-making processes demonstrates the practicability of the proposed model and method in social interactive systems.
Article
Predicting the winner of an election and estimating the margin of victory of that election are favorite problems both for news media pundits and computational social choice theorists. Since it is often infeasible to elicit the preferences of all the voters in a typical prediction scenario, a common algorithm used for predicting the winner and estimating the margin of victory is to run the election on a small sample of randomly chosen votes and predict accordingly. We analyze the performance of this algorithm for many commonly used voting rules. More formally, for predicting the winner of an election, we introduce the (ε,δ)-Winner Determination problem, where given an election E on n voters and m candidates in which the margin of victory is at least εn votes, the goal is to determine the winner with probability at least 1−δ where ε and δ are parameters with 0<ε,δ<1. The margin of victory of an election is the smallest number of votes that need to be modified in order to change the election winner. We show interesting lower and upper bounds on the number of samples needed to solve the (ε,δ)-Winner Determination problem for many common voting rules, including all scoring rules, approval, maximin, Copeland, Bucklin, plurality with runoff, and single transferable vote. Moreover, the lower and upper bounds match for many common voting rules up to constant factors. For estimating the margin of victory of an election, we introduce the (c,ε,δ)–Margin of Victory problem, where given an election E on n voters, the goal is to estimate the margin of victory M(E) of E within an additive error of cM(E)+εn with probability of error at most δ where ε,δ, and c are the parameters with 0<ε,δ<1 and c>0. We exhibit interesting bounds on the sample complexity of the (c,ε,δ)–Margin of Victory problem for many commonly used voting rules including all scoring rules, approval, Bucklin, maximin, and Copelandα. We observe that even for the voting rules for which computing the margin of victory is NP-hard, there may exist efficient sampling based algorithms for estimating the margin of victory, as observed in the cases of maximin and Copelandα voting rules.
Chapter
The main objective of the chapter is to provide an insight into the motivation mechanisms for the crowd to participate in crowdsourcing projects. For this to happen, the authors investigate the factors which have been suggested in the literature as influencing the motivation of the crowd and the task type in each study in the related literature and contrasted the motivation factors in various contexts. The systematic literature review method has been used for the purpose of this study. This involved a comprehensive search in five scientific databases which resulted in 575 papers. This initial pool of studies has been refined in various rounds and ended in identification of 37 studies which directly targeted the topic of motivation in crowdsourcing. The study introduces various categories of motivations and investigates the factors which have been utilized in each context. Finally, possible implications for practice and potential research gaps are discussed.
Chapter
This chapter proposes concepts for designing and developing decision support systems that acknowledge, explore and exploit the fact that conversations among people are the top-level “supporting device” for decision-making. The goal is to design systems that support, configure and induce increasingly effective and efficient decision-making conversations. This includes allowing and motivating participation in decision-making conversations of any people who may contribute positively to decision-making and to the quality of its outcomes. The proposal sees the sum total of decisions being taken in an organization as the global decision process of the organization. The global decision process of the organization is structured in decision processes corresponding to organizational domains. Each organizational domain has associated a unit decision process. If the organizational domain contains organizational sub-domains, then its compound decision process is the union and composition of its unit decision process and the unit decision processes of its sub-domains. The proposal can be seen as extending, enlarging and integrating group decision support systems into an organization-wide system. The resulting organizational decision support system, by its conversational nature, may become the kernel decision support system of an organization or enterprise. In this way, the global decision process of the organization may be made explicit and monitored. It is believed that this proposal is original.
Chapter
This chapter proposes concepts for designing and developing decision support systems that acknowledge, explore and exploit the fact that conversations among people are the top-level “supporting device” for decision-making. The goal is to design systems that support, configure and induce increasingly effective and efficient decision-making conversations. This includes allowing and motivating participation in decision-making conversations of any people who may contribute positively to decision-making and to the quality of its outcomes. The proposal sees the sum total of decisions being taken in an organization as the global decision process of the organization. The global decision process of the organization is structured in decision processes corresponding to organizational domains. Each organizational domain has associated a unit decision process. If the organizational domain contains organizational sub-domains, then its compound decision process is the union and composition of its unit decision process and the unit decision processes of its sub-domains. The proposal can be seen as extending, enlarging and integrating group decision support systems into an organization-wide system. The resulting organizational decision support system, by its conversational nature, may become the kernel decision support system of an organization or enterprise. In this way, the global decision process of the organization may be made explicit and monitored. It is believed that this proposal is original.
Article
Full-text available
A recent work (Hernández, et al., 2018) introduced a networked voting rule supported by a trust-based social network, where indications of possible representatives were based on individuals opinions. Individual contributions went beyond a simple vote-counting and were based on proxy voting. This mechanism selects committees with high levels of representativeness, weakening the possibility of patronage relations. By incorporating the integrity of individuals and its perception, we here address the question of the resulting committee’s trustability. Our results show that this voting rule provides sufficiently small committees with high levels of representativeness and integrity. Furthermore, the voting system displays robustness to strategic and untruthful application of the voting algorithm.
Article
Approval voting provides an opportunity for the agents to make a comment about every candidate, without incurring the overhead of determining a full ranking on the entire set of candidates. This makes approval voting a natural choice for many practical applications. In this work, we focus on the use of approval voting for selecting a committee in scenarios where we can have few outrageous voters whom we call outliers. More specifically, we study the computational complexity of the committee selection problem for commonly used approval-based voting rules in the presence of outliers. Our first result shows that outliers render the committee selection problem intractable for approval, net approval, and minisum approval voting rules. We next study the parameterized complexity of this problem with five natural parameters, namely the target score, the size of the committee (and its dual parameter namely the number of candidates outside the committee); and the number of outliers (and its dual parameter namely the number of non-outliers). Our main contribution in this paper is dichotomous results, which establish the parameterized complexity of the problem of selecting a committee under approval, net approval, and minisum approval voting rules for all subsets of the above five parameters considered here (by showing either FPT or W[1] -hardness for all subsets of parameters).
Article
Full-text available
This thesis is in the area called computational social choice which is an intersection area of algorithms and social choice theory.
Article
Full-text available
Past evidence has shown that collaborative filtering approaches to recommender systems tend to poorly scale. Moreover, they have limited fitness for scenarios with distributed data storage and decentralized control, like the Semantic Web. We believe that computa-tional trust models bear several favorable properties for social filtering, which opens new opportunities by either replacing or supplementing current techniques. However, in order to provide meaningful results for recommender system applications, we expect notions of trust to clearly reflect user similarity. The study at hand provides empirical results ob-tained from two real, operational communities and verifies the latter hypothesis for the domain of book and movie recommendations.
Conference Paper
Full-text available
This paper proposes a new type of Information/Communication system or Inquiry System to support the investigation by large groups (e.g. thousands of people) of complex topics about which many diverse and opposing views are held. The objective of this system is the production, integration, and synthesis of diverse views in a manner in which: (a) all participants can come to respect and understand the differences caused by diverse values and interests of the contributing population, and (b) there can be a movement towards consensus on at least some of the issues involved. Controversies related to Pervasive Information Technology are used as an example, and challenges to participation raised by the "digital divide" are discussed.
Article
This paper presents a general purpose framework for the design and implementation of a large scale collective decision making system. The unifying construct is a directed, weighted semantic network that connects humans, their mental models, and their artifacts to enable swarms of particles to traverse the network and rank solutions to the problems facing the group.
Article
Intelligent Information Retrieval is concerned with the application of intelligent techniques, like for example semantic networks, neural networks and inference nets to Information Retrieval. This field of research has seen a number of applications of Constrained Spreading Activation (CSA) techniques on domain knowledge networks. However, there has never been any application of these techniques to the World Wide Web. The Web is a very important information resource, but users find that looking for a relevant piece of information in the Web can be like ‘looking for a needle in a haystack’. We were therefore motivated to design and develop a prototype system, WebSCSA (Web Search by CSA), that applied a CSA technique to retrieve information from the Web using an ostensive approach to querying similar to query-by-example. In this paper we describe the system and its underlying model. Furthermore, we report on an experiment carried out with human subjects to evaluate the effectiveness of WebSCSA. We tested whether WebSCSA improves retrieval of relevant information on top of Web search engines results and how well WebSCSA serves as an agent browser for the user. The results of the experiments are promising, and show that there is much potential for further research on the use of CSA techniques to search the Web.
Article
Online communities that allow their users to express their personal preferences, such as the members they trust and the products they appreciate, are becoming increasingly popular. Exploiting these communities as playgrounds for sociological research we present two frameworks for analyzing the correlation between interpersonal trust and interest similarity. We obtain empirical results from applying the two frameworks on two real, operational communities, that suggest there is a strong correlation between both trust and interest similarity. We believe our findings particularly relevant for ongoing research in recommender systems and collaborative filtering, where people are suggested products based on their similarity with other customers, and propose ways in which trust models can be integrated into these systems.
Article
Collective Intelligence (CI) is the ability of a group to exhibit greater intelligence than its individual members. Expressed by the common saying that "two minds are better than one," CI has been a topic of interest for social psychology and the information sciences. Computer mediation adds a new element in the form of distributed networks and group support systems. These facilitate highly organized group activities that were all but impossible before computer mediation. This paper presents experimental findings on group problem solving where a distributed software system automatically integrates input from many humans. In order to quantify Collective Intelligence, we compare the performance of groups to individuals when solving a mathematically formalized problem. This study shows that groups can outperform individuals on difficult but not easy problems, though groups are slower to produce solutions. The subjects are 57 university students. The task is the 8-Puzzle sliding tile game.
Article
Will Internet-based haptics have the right touch?