This paper presents a method for building concept lattices by learning concepts from RDF annotations of Web documents. It consists in extracting conceptual descriptions of the Web resources from the RDF graph gathering all the resource annotations and then forming concepts from all possible subsets of resources-each such subset being associated with a set of descriptions shared by the resources belonging to it. The concept hierarchy is the concept lattice built upon a context built from the power context family representing the RDF graph. In the framework of the CoMMA European IST project dedicated to ontologyguided Information Retrieval in a corporate memory, the hierarchy of the so learned concepts will enrich the ontology of primitive concepts, organize the documents of the organization's Intranet and then improve Information Retrieval. The RDF Model is close to the Simple Conceptual Graph Model; our method can be thus generalized to Simple Conceptual Graphs.