Tabular format is a common format in open data. However, the meaning of columns is not always explicit which makes if difficult for non-domain experts to reuse the data. While most efforts in making data FAIR are limited to semantic metadata describing the overall features of datasets, such a description is not enough to ensure data interoperability and reusability. This paper proposes to reduce this weakness thanks to a (FAIR) core semantic model that is able to represent different kinds of metadata, including the data schema and the internal structure of a dataset. This model can then be linked to domain-specific definitions to provide domain understanding to data consumers .