In order to select self-medication relevant molecules from our LOD data sets, our first approach was based on identifying molecules with the ATC or EphMRA classifications. We rapidly found out that these identifiers are sparsely used in DrugBank and DBPedia, the central data sets of our application. So we complemented our ontology with a new identifier solution based on DrugBank keys. This processing step was performed semi-automatically using different sources such as DBPedia and other classifications, e.g., EphMRA. At this stage, we also found out that some molecules could not be matched because they could not be found due to their non-commercialization in North America, e.g., absence of the Ambroxol molecule. We found out that most missing drugs were badly rated and considered useless in our French system.