![](bg2.png)
service providers as long as that user’s trust in the new
provider is sufficiently high.
It is not obvious that adding global considerations
without degrading the user’s service is possible. To
the best of our knowledge no such manipulations were
considered in previous works in the field of recom-
mendation systems. Other fields of research, on the
other hand, deal with somewhat similar problems. For
example, excessive waiting by users in a medical sce-
nario (Silvester et al., 2004) such as the one described
above. Questions relating to the capacity of health
organizations and the waiting queue of patients are
examined, but the main focus is the adaptation of an
organization’s management system to accommodate
patients (users) flow through them.
A reputation-based QoS estimation approach has
been studied in (Vu et al., 2005). The QoS-based
web service selection and rating algorithm presented
in this paper, returns a list of services meeting qual-
ity criteria set by the requesting user. Users reports
on QoS of all services are collected over time to pre-
dict their future quality. This prediction is also based
on the quality promised by the service providers and
considers trust and reputation issues as well. The au-
thors demonstrate the efficiency of the algorithms un-
der various cheating behaviors of users. A later study
presented a framework for the autonomous discovery
of semantic web services based on their QoS proper-
ties (Vu et al., ). The framework addresses various
aspects such as semantic modeling of QoS, personal-
ized matchmaking and rating of services, and the use
of services QoS reputation in the discovery process.
However, non of these papers attempt to manipulate
the rating for achieving a community-wide benefit.
An interesting approach uses Game Theory to in-
troduce an interested party which may not enforce be-
havior and payments or redesign the system (Mon-
derer and Tennenholtz, 2004). The interested party
directs users’ behavior by committing to non negative
monetary transfers, and it is not clear that this is pos-
sible in our scenario.
The remainder of this paper deals with a proposed
method for the introduction of the system’s interests.
Before any administrator intervention can take place,
each user should provide a service level threshold.
As long as this condition is met, the user will not be
harmed by the changes. This provides a motivational
basis for the system, often termed Individual Ratio-
nality.
Our ideas are applied to a virtual community of
users and service providers. Members of the commu-
nity interact with each other based on the trust level
they have in each other. When a member desires some
service, she queries the system for it. The system out-
puts a recommendation in the form of an ordered list
of service providers matching the query and the user,
along with their respective ratings. These ratings rep-
resent the aggregated rating of the service providers
with respect to the user and her trust level in her peers.
Our proposed approach manipulates the list of rec-
ommended services. For example, a service provider
can be promoted or demoted according to her avail-
ability. This effectively shortens waiting queues. In
a different scenario one may wish to manipulate the
services list according to seniority. This may better
reflect notions of “hospitality” or “hostility” toward
new service providers.
The proposed manipulation is by no means mis-
leading to the users or service providers. All partici-
pants must agree to this form of manipulation before
joining the service. Moreover, users explicitly pro-
vide the level of manipulation threshold they are will-
ing to accept.
2 A THRESHOLD FOR SERVICE
PROVIDERS
Let us first define the condition for satisfaction of a
service provider. The gain of a service provider from
using the system is a sum of all gains from every inter-
action with a user. When the service provider is over
burdened with incoming requests, not all requests will
be treated by the service provider within a given time
unit. These requests will be processed at a later time.
More formally, we use the following notation:
Definition 1. Queue size - The queue size of a service
provider e at a given Time Unit (TU) t is e
q
(t).
Definition 2. Capacity - A service provider e can han-
dle e
c
requests in a given TU.
Definition 3. Payment - The gain of a service
provider e from processing a single user request is e
p
.
The total profit of e is the sum of payments made
to her. This means that the profit of a service provider
is only dependent upon the number of requests she
handles. As mentioned before this profit is limited
due to capacity constraints: in a given TU, a service
provider may handle up to e
c
requests from her queue.
Hence, the profit of a service provider e is:
e
pro f it
(t) =
e
p
· e
q
(t) if her queue size is not
larger than her capacity,
e
p
· e
c
otherwise.
(1)
Following Eq. 1 we conclude that a service
provider will not be harmed by the system, as long
ICAART 2010 - 2nd International Conference on Agents and Artificial Intelligence
136