Computing similarities between Web services is a main concern in Service Oriented Architecture as it allows to decide which services are likely to be matched into a composite workow, or in other cases, which services can be substituted in order to ensure continuous service availability. With the high maturity achieved by the standards, tools and frameworks in the Semantic Web domain, measuring Web services similarities relies more than ever on semantic descriptions of services as well as on semantic relationships these descriptions may hold. In this paper we present a Framework for Web services composition based on computing semantic similarity between Web services. We particularly focus on Services Matching engine which uses the considered similarity measure first to classify Web services into classes of functionally similarWeb services and then to propose a composite sequence of services that matches a requested goal. In both tasks, the presented framework appeals for best known techniques of similarity computing and data and knowledge extraction, respectively.