Abstract

The notion of community of web services has been recently proposed and investigated to gather functionally similar web services in the same virtual space. This allows
increasing the visibility of web services and their collaboration, which makes their discovery and composition easier. Using the community infrastructure, users are supposed
to direct their requests to the community's manager (called master), that is in charge of
selecting the appropriate web service. Because many communities providing the same
functionality are available, selecting the best community to deal with, from the users
and providers perspectives, is a key factor that still needs to be investigated. Another
particularly challenging issue yet to be addressed is the selection by the master of the
appropriate web service to be hosted in the community. Reputation has been proposed
as a means to help users, providers, and masters evaluate and rank different candidates.
However, reputation is mainly based on users feedback, which is not always accurate.
Moreover, other performance parameters should be considered in the selection game.
In this thesis, we propose a new assessment process that focuses on various performance aspects of the community rather than just its reputation. This assessment
considers the performance parameters from the users, providers, and masters perspectives. In this approach, the communities performance rate is mainly based on the web
services hosted by those communities. Such an assessment approach helps the master of
the community differentiate between web services so that only the appropriate ones can
be invited or accepted to join based on the communities requirements. It also helps the
users and providers select the best available communities.
The proposed method works on three steps. The first step focuses on defining and
iv
computing the evaluation metrics used in the assessment process while considering the
requirements of all the stakeholders, namely users, providers, and communities. Thus,
each community or web service is described by a vector of metrics. The second step
includes the clustering of the evaluated communities and web services using the resulted
vectors from the first step. During the third step, the resulting clusters are ranked using
a function called goodness function. Web services and communities belonging to the
best cluster are then selected. The effectiveness of the proposed assessment approach is
tested by simulation and comparison to two other approaches in the literature.