Trust and Technology - Trust Theory: A Socio-Cognitive and Computational Model - page 345

We have to say that each of these metrics can present both advantages and problems, but
what is important in our view is the fact that trust has an intrinsic multi-factorial nature and
this peculiar feature has to be represented and implemented also in the quantitative aspects of
this phenomenon (see Chapter 3).
One of the more interesting attempts in this direction is represented by the REGRET
approach where to overcome the mono-dimensionality of the trust performance some fuzziness
over the notion of performance itself is used. REGRET introduces a rich semantics for the
ratings (called impressions ) by defining their specific features (for example: delivery date ,
price , and so on). On the basis of the fuzzy reasoning techniques, the system is able to
compose the different dimensions producing a general and global impression of one agent
on another. With respect to the fuzzy approach applied to the trust evaluation we showed in
Chapter 11 a specific implementation based on the socio-cognitive approach to trust.
Other Models and Approaches to Trust in the Computational Framework
Recent works on designing models for propagating trust in social networks and for selecting
the most appropriate services in the service-oriented environments are of particular interest to
analyze. In (Yu and Singh, 2003) the models of trust propagation are distinguished by:
Trust in expertise : ability of providing services; and
Trust in sociability : ability of providing referrals.
These two functions are quite different even if relevant: the former is becoming more
important given the increasing use of the Internet by people (Internet users are 'now', 40 years
after its birth, 1,663 million) and its circulation of both traditional and innovative services.
With regard to the latter, if we use trust to evaluate people and information we have
to compute trust between people who do not know one another and expect to achieve the
result that each agent in the network will evaluate the trustworthiness of its potential, often
anonymous, partners (Golbeck, 2009).
In CertProp model (Hang et al. , 2009) i a trust propagation model based on three operators
( aggregation , concatenation and selection ) s introduced to efficiently and accurately propagate
trust in social networks. In this approach a social network (system of interacting agents) is
modeled as a directed graph with weights on the links: a sort of social graph , in which the
nodes represent the users and the edges represent the relationships between these users (the
social graph representation is quite widespread and used, see also (Ziegler, 2009), (Levien,
2009)). The problem of propagating trust in this network using the introduced operators is
interesting even if not all the problems connected with the trust propagation are solved: the
already presented question about the so-called 'trust transitivity' (see Chapter 6) remains an
issue that has not been well addressed.
In any case this work, defining a set of algebric properties for the three operators, determines
trust propagation in an efficient and accurate way (even if in a simplified domain). Another
interesting question analyzed in this work is the classical mapping between opinions and
evidences : in the networks the weights are subjective opinions and not objective evidence.
Then, the authors propose approaches for transforming opinions in evidence: in particular
they motivate a new way of this transformation based on Weber fechner law (that describes