You are on page 1of 4

Bayesian Belief Network Models of Trust and Social Capital for Social

Software Systems Design

Ben K. Daniel1, 3, Juan-Diego Zapata-Rivera2, Richard A. Schwier3 & Gordon I. McCalla1

ARIES Research Group1


Department of Computer Science
University of Saskatchewan
Canada
Educational Testing Services2
USA
Educational Communications and Technology3
University of Saskatchewan, Canada

ABSTRACT mutual understanding, and shared values and


BBN models enable reasoning when there is behaviors that bind the members of communities
uncertainty. They combine the advantages of an together. In the context of virtual communities,
intuitive visual representation with a solid positive interaction provides value to its
mathematical basis in Bayesian probability. This
participants especially when it is built upon
paper builds a computational framework for modeling
imprecise and inconsistent data that is relevant to positive attitudes among individuals in a
social software systems design. Drawing from the community [2]. Putnam [10] suggested that
framework, we illustrate the process involved in building social capital requires continuous and
building Bayesian belief models of trust and social positive interaction.
capital in virtual communities.
Further, positive interaction occurs when
INTRODUCTION participants have a common set of expectations,
Bayesian Belief Network (BBN) techniques are mediated by a set of shared social protocols and
increasingly being used for understanding and shared understanding [2]. Furthermore, different
simulating computational models of social forms of awareness in a virtual community can
software systems. BBN models enable reasoning encourage positive interaction and productive
when there is uncertainty [8]. They combine the collaboration [6][7]. Awareness can also foster
advantages of an intuitive visual representation trust; trust develops as people become
with a sound mathematical basis in Bayesian sufficiently aware of others in the community
probability. Though BBN techniques are elegant (who is who, who knows who, who knows what,
ways for capturing uncertainties, considerable who is located where) and learn what to expect
effort is required to create conditional probability from each other. Being able to examine others’
values for the given variables in a network. We past behaviors in the community is necessary for
propose a computational framework drawing building trust [9] [1].
from both qualitative and quantitative
descriptions of conditional probability values, BAYESIAN BELIEF NETWORK
and we illustrate how the framework has been A Bayesian network is a particular type of
used to build models of trust and social capital graphical model, frequently used in applications
that can be used as decision tools for the design of artificial intelligence for building probabilistic
of social software systems. expert systems. Bayesian networks can be used
to model probabilistic relationships among
SETTING THE MOEDELLING SCENE variables. In some cases, their graphical structure
Social capital is one useful framework for can be loosely interpreted as the result of direct
addressing social issues in communities and causal dependencies between variables. In
distributed groups; however, the concept remains domains with many causal relations, such as in
elusive and imprecise. Many variables comprise medical diagnosis (symptoms cause diseases),
the construct of social capital. Cohen and Prusak human experts are usually able to express their
[1] noted that social capital consists of the stock domain knowledge in the graphical structure of
of active connections among people, the trust,
the network. For example, in a model for identification of possible critical variables in a
medical diagnosis, the parameters of the network particular domain, along with their possible
are the conditional probabilities of effects given states [5]. In our work we have identified
the state of their direct causes. variables constituting social capital drawn
mainly from the literature (see table1) and added
Constructing a BBN generally involves several others that we believe may be relevant in the
steps. The first step in constructing BBN models context of virtual. The variables identified are
is to define the problem or opportunity that needs then assigned potential states.
to be modeled. This is followed by the

Table 1. Operationalisation of variables of social capital

Name Definition States

Interaction Exchanging of information Present/Absent


between two or more
individuals via text, video, or
any other digital media
Attitudes Individuals' general Positive/Negative
perception about each other
and others' actions
Community The type of environment, Virtual learning community
Type tools, goals, and tasks that (VLC) or Distributed
define the group community of practice
(DCoP)

Shared A mutual High/Low


Understanding agreement/consensus
between two or more agents
about the meaning of an
object
Awareness Knowledge of people, tasks, Present/Absent
or environment and or all of
the above
Demographic Knowledge of an individual: Present/Absent
Awareness country of origin, language
and location
Professional Knowledge of an individual’s Present/Absent
Awareness background training,
affiliation etc.
Competence Knowledge about an Present/Absent
Awareness individual’s capabilities,
competencies, and skills
Capability Knowledge of an individual’s Present/Absent
Awareness competences and skills in
regards to performing a
particular task
Social The mutually agreed upon, Present/Absent
Protocols acceptable and unacceptable
patterns of behaviour in a
community
Trust A particular level of certainty High/Low
or confidence with which an
agent uses to assess the action
of another agent
and positive interactions. Using this scenario to
The second step in building a BBN model query the network, we observed higher level of
involves mapping identified variables to a trust P (Trust=0.93) and subsequently, increasing
network structure based on logical, and coherent level of social capital P (Social capital=0.73)
qualitative reasoning. The resulting acyclic (see Figure 2).
directed graph shows dependencies among
variables (see figure 1).

Figure 2. High shared understanding, awareness


Figure 1. A BBN model of trust and social capital and positive interactions
in virtual communities [3]
By comparison, we consider the level of social
The third step in building the model involves capital and trust in a distributed community of
assigning initial probabilities to each node in a practice (DCoP), whose members have low level
network. In our approach initial prior and of shared understanding, and who are not aware
conditional probabilities are generated by of their professional affiliations, but who
qualitative descriptions of the strength of the nevertheless have come together to interact
relationship among variables in a network. This around certain issues. Updating the network
approach takes into account the number of states reveals significant drop in probabilities of trust P
of a variable, the number of parents, the relative (Trust=0.73) by 0.2 and social capital P (Social
strength of a variable (e.g., strong -S, medium - capital = 0.6) by 0.13 (see Figure 3).
M, weak -W) and the kind of
relationship/influence of the variable (e.g.,
positive or negative influence - +/- ) to produce
initial prior and conditional probabilities. Once
an initial model is elicited, particular scenarios
are used to refine and document the network [3].

QUERYING THE NETWORK


The fourth step in the modeling process is to
observe changes in the network when new
evidence affecting one or more of the nodes is
added to the network. Querying a BBN refers to
the process of updating conditional probability
tables and making inferences based on new Figure 3. Low shared understanding and absence
evidence. One way of analyzing and refining a of awareness
BBN is to develop a detailed number of
scenarios or cases grounded on data, intuition, The two scenarios presented above were
and experts’ opinions describing a set of developed with the assumptions that a distributed
phenomena in a virtual community. A community of practice is typically a group of
scenario/case refers to a written synopsis of geographically dispersed professionals in
inferences drawn from observed phenomena, different fields who share common practices and
intuition or empirical data. For illustration, we interests in a particular area of concern, and
describe a scenario in a distributed community of whose activities can be enriched and mediated by
practice, whose members have a high level of information and communication technologies.
shared understanding, professional awareness Such a group usually maintains high level of
shared understanding and professional awareness, 2. Daniel, B.K., McCalla, G. I. &
and so variation in the level of shared Schwier, R. A. Social capital in virtual
understanding and awareness of each other learning communities and distributed
within such a group can affect the level of trust communities of practice. The
and social capital as demonstrated by changes in Canadian Journal of Learning
the probability distributions in the network. Technology, 29(3), (2003), 113-139.
3. Daniel, B.K., Zapata-Rivera, J. D., &
CONCLUSION McCalla, G. I. A Bayesian
Bayesian networks techniques are well suited to computational model of social capital
handle partial knowledge of a domain and in virtual communities. In M.
effectively make coherent predictions, and they Huysman, E. Wenger, & Wulf, V.
provide a natural way to encode dependencies (eds), Communities and technologies
among variables in a domain. The approach (2003), 287-305. London: Kluwer.
presented in this paper allows researchers to 4. Daniel, B.K., McCalla, G.I., &
easily build sophisticated initial computational Schwier, R.A. Data mining and
models based on experts’ opinions and refine the modeling social capital in virtual
models as new evidence is gathered. It also learning communities. Proceedings of
enables them to perform model verifications the 12th International Conference on
using pre-post sensitivity analysis, to establish Artificial Intelligence in Education,
the strength of casual relations among variables Amsterdam, 18-22 July, (2005 ), 200-
in the network, as well as corroborate results 208
with other techniques to reach valid, coherent 5. Druzdzel, M.J., & Gaag, L. C.
and reliable conclusions. Building probabilistic networks:
"Where do the numbers come from?"
Scenarios/cases for updating a network as shown Guest editor's introduction. Data
in the two examples can provide an effective Engineering, 12(4), (2000), 481-486.
way of reasoning when there is uncertainty in the 6. Gutwin, C. and Greenberg, S. Effects
system and in domains where there are no of awareness support on groupware
concrete data to base judgments or decisions. usability. Proceedings of the ACM
Predictions based on the two scenarios show that Conference on Human Factors in
trust in virtual communities can be influenced by Computing Systems, (1998), 511-518.
variables such as awareness, social protocols and 7. Ogata, H. & Yano, Y. Combining
shared understanding. The predictions also knowledge awareness and information
suggest that trust is context dependent on the filtering in an open-ended
nature of the community and individuals in that collaborative learning environment.
community. International Journal of Artificial
Intelligence in Education, 11, (2000),
Though we have not conducted specific 33-46
experiments correlating trust with different kinds 8. Pearl, J. Probabilistic reasoning in
of awareness as shown in the model, we suggest intelligent systems: Networks of
our approach sets the stage for such studies. It is plausible inference. San Mateo, CA:
also possible to extend this framework to model Morgan Kaufmann, (1988).
similar problems in other domains relevant for 9. Preece, J. Supporting community and
building social software. And finally, the building social capital. Special edition
combination of qualitative and quantitative of Communications of the ACM, 45, 4
reasoning can support both hard and soft data (2002), 37- 39.
structures needed for building Bayesian network 10. Putnam, R. Bowling alone: The
models. collapse and revival of american
community, New York: Simon
REFERENCES
Schuster, (2000).
1. Cohen, D. & Prusak, L. In good 11. Resnick, P. Beyond bowling together:
company: How social capital makes Sociotechnical capital. In J. M. Carroll
organizations work. Massachusetts: (Ed), HCI in the new millennium,
Harvard Business School Press, (2002), 247-272. New York: Addison-
(2001). Wesley.

You might also like