|
With the increased significance of the Internet in our everyday
lifes, we embrace its benefits as seemingly unlimited information
source, warehouse and general communication medium, but sometimes
fall prey to its predators. Outside the online world, social network
structures of friends or colleagues allow to identify malicious and
reputable entities and to communicate recommendations or warnings
accordingly. When interacting through open computer networks, these
traditional mechanisms used in the physical world for establishing
trust are adapted by reputation systems that allow to build trust in
entities and create social network structures on a much larger
scale.
In this dissertation, we investigate various models and algorithms
required for realizing a fully decentralized reputation system with
enhanced privacy properties and fine-grained trust modeling. To
ensure the former, we bind trust to virtual identities instead of
real identities and present extended destination routing, an
approach that allows anonymous communication between pseudonyms
without exposing any link to a real identity. To enable the latter,
we introduce a generic trust model that allows to model trust in
various context areas in addition to expressing context area
dependencies that are taken into account when updating trust. The
model definition permits incorporating several well-known trust
update algorithms from the related work. Subjecting the algorithms
to a set of evaluation scenarios gives valuable inputs regarding
their specific performance. In order to capture the transitivity of
trust, we present algorithms to simplify trust networks and then
compute the transitive trust with subjective logic operators.
Finally, we propose mechanisms to protect trust by firstly laying
its foundation in trusted hardware and secondly ensuring the
authenticity of recommendations through the integration of an
originality statement.
This reputation system can be utilized by users and relying
applications alike to determine the trustworthiness of other
entities. While these building blocks are all essential for our
system, many contributions can be applied to other reputation
systems and even to other research areas as well.
|