The RDF data sources published on the Web represent an unprecedented amount of knowledge. However, querying these sources to extract the relevant information for some specific needs represented by a target schema is a complex task, as the alignment between the target and the source schemas might not be provided or may be incomplete. This paper presents a system that aims to automatically populate the classes of a target schema from RDF data sources by identifying candidate instance patterns. This identification process relies on a semi-supervised learning algorithm and the system automatically generates the SPARQL queries that populate the target schema.