Trust and Technology - Trust Theory: A Socio-Cognitive and Computational Model - page 347

Another interesting work about trust propagation in social networks is focused on a different
aspect of trust inference: in particular, the change in trust values in the network and the impact
of that change on the results of the existing algorithms. In their work (Golbeck and Kuter,
2009), Golbeck and Kuter show an experimental study to understand the behavior of different
trust inference algorithms with respect to the changes occuring in the social network. Their
contribution is on different relevant items that they define with the following questions:
'How far does a single change propagate through the network? How large is the impact of
that change? How does this relate to the type of inference algorithm?' ((Golbeck and Kuter,
2009), p. 170). They show the relevance of the chosen algorithm in all three questions.
As we have claimed above the problem of selecting trustworthy services on the web is
becoming really relevant given the variety and quantity of the offer. In particular, two main
problems should be taken in consideration:
a) The trust evaluation of a service should take into consideration the fact that very often a
service is a composed service (with different providers, functions and responsibilities).
b) The dynamism of service providers and consumers (both the needs of the consumers and
the providers' quality of service, continuously change).
Singh (Singh, 2003), has coined the term ' service-oriented computing ' for meaning 'the
emergence of independent services that can be put together dynamically at run time and
possibly across administrative domains' (p.39). In fact, while works exist about modeling
trustworthiness of individual service providers, very few results were reached in modeling
groups of providers working to a specific composed service. This approach is also important
because - like in our model - consider the 'quality' or the 'competence' intrinsic part of
trustwortiness and of trust; differently from many models (in game theory, in philosophy,
etc.) that want to eliminate this component/dimension of trust to reduce it only to honesty,
reliability, morality, etc.
Hafizoglu and Yolum (Hafizoglu and Yolum, 2009), propose a group trust model to under-
stand the behavior of such teams that realize a composed service. Their work is in turn based
on the service graphs model (Yolum and Singh, 2004) where graphs are helpful for reasoning
about services that are related to each other. In (Hafizoglu and Yolum, 2009), the authors
have to cope with the problem that 'the behavior of an agent in teamwork environment may
differ from its behavior in single service environment'. In fact, collaboration may have some
influence on the agents' performances. So the individual features of the agents in providing
specific services are not so useful for selecting the right (best) team of the composed service.
The authors analyze a set of possible tendencies for the agents ( ideal behavior , group antipa-
thy , group motivation , colleague effect , task effect , familiarity effect ) that have influence on
the agents' collaborative performances.
Another interesting work on this problem is (Hang and Singh, 2009) in which is proposed
a trust-aware service selection model based on a Bayessan network. The model evaluates the
service trustworthiness on both direct and indirect (from referrals) experience. The method
models causal relationships between services with Bayesan networks.
The main characteristic of this model is that it can deal with incomplete observation (in
such a way taking into account the possibility that underlying services may not be exposed to
the consumer) for this introduces a specific parameter representing the percentage of missing
data.