Trust and Technology - Trust Theory: A Socio-Cognitive and Computational Model - page 342

study field; in fact, on the basis of reputation a set of automatic models and systems for
attributing trust were directly studied and built in the last 15 years. In many cases the reputa-
tional approach represents the only criterion for defining the trustworthiness of the interactive
agents.
Reputation mechanisms are distinguished in centralized and decentralized mechanisms.

Centralized Reputation Mechanisms

Centralized reputation mechanisms are widespread in electronic commerce: eBay [eBay site]
(Resnick and Zeckhauser, 2002), Amazon (Amazon site], and many others' e-commerce sys-
tems manage these kinds of mechanisms in which all the users have their own reputational
profile stored in a centralized database. In these systems each user, after an interaction (trans-
action) with other users, reports on the behavior of the other providing appropriate ratings and
giving textual comments. These ratings and comments are public and each user can read them
before starting a new interaction/business with a specific agent. In the eBay system the rate
scale is from 1 (respectively negative, neutral and positive). All the ratings are stored

centrally and the global reputation value is the sum of all the ratings in the last six months. The
main limit of this approach is given by the extreme simplicity of the model. In fact, just one
dimension of the trustworthiness is taken in consideration (that is a more complex entity) and
the acritical aggregation of the performances do not give account of the possibility of cheating
in few interactions maintaining a good reputation value.
To overcome the limits of the reputation systems shown above, SPORAS (Zacharia and
Maes, 2000) has been developed and it introduces new methods for aggregating the ratings.
In particular, the updating of the ratings follows these principles:

− 1, 0, +

1. New users start with a minimum reputation value and they build up reputation during their
activity on the system.
2. The reputation value of a user never falls below the reputation of a new user;
3. After each transaction, the reputation values of the involved users are updated according to
the feedback provided by the other parties, which reflect their trustworthiness in the latest
transaction;
4. If two users happen to interact more than once the system keeps the most recently submitted
rating.
5. Users with a very high reputation value experience much smaller rating changes after each
update.
6. Ratings must be discounted over time so that the most recent ratings have more weight in
the evaluation of a user's reputation.

The six above principles define a more interesting dynamics of the reputation model with
respect to more static ones like eBay ,or Amazon . In addition, SPORAS introduces a measure
of the reliability of the users' reputations: This reliability is based on the deviation of rating
values. In this way this system introduces an indication of the predictive power of the algorithm.
High deviations correspond to high degrees of variation (or to insufficient activation in the
transactions).