This paper describes structures which represent the organization of documents extracted from Open Archives Initiative compliant data providers. We have called them "ontologies of records". They group similar documents by means of data mining techniques and document clustering algorithms. Ontology markup languages are used to implement them. We show an ontology of records constructed in a semi-automatic way and propose a maintenance process based on collaborative rewriting and revision. Ontologies of records have a well defined meaning, they enable human and software agents to work in cooperation to exploit data providers. The paper is a small contribution to the construction of lightweight ontologies to be used for different purposes in the semantic web.