The success of the Web service technology has brought topics such as software reuse and discovery once again on the agenda of software engineers. While there are several efforts towards automating Web service discovery and composition, many developers still search for services via online Web service repositories and then combine them manually. However, from our analysis of these repositories, it yields that, unlike traditional software libraries, they rely on little metadata to support service discovery. We believe that the major cause is the difficulty of automatically deriving metadata that would describe rapidly changing Web service collections. In this paper, we discuss the major shortcomings of current Web service repositories and, as a solution, we report on how to use techniques developed in the context of the Semantic Web (ontology learning, service annotation tools, automatic classifiers, metadata based presentation) to improve the current situation.