Linked Open Data (LOD) constitute a large and growing collection of inter-domain data sets. LOD are represented as RDF graphs that allow interlinking with ontologies, facilitating data integration, knowledge engineering and in a certain sense knowledge discovery. However, ontologies associated with LOD are of different quality and not necessarily adapted to all data sets under study. In this paper, we propose an original approach, based on Formal Concept Analysis (FCA), which builds an optimal lattice-based structure for classifying RDF resources w.r.t. their predicates. We introduce the notion of lattice annotation, which enables comparing our classification with an ontology schema, to confirm subsumption axioms or suggest new ones. We conducted experiments on the DBpedia data set and its domain ontologies, DBpedia On-tology and YAGO. Results show that our approach is well-founded and illustrates the ability of FCA to guide a possible structuring of LOD.