With the increasing use of the Web of Data and the development of data based analysis applications, there is a real need for developing suitable methods and techniques for evaluating and help ensuring web data quality. Among Quality dimensions, completeness is recognized as difficult to evaluate, as it often relies on gold standards and/or a reference data schema that are neither always available nor realistic from a practical point of view. In this paper, we propose an approach for Linked Data completeness assessment. The approach is a two-step process: first, mining from a data source, a schema that reflects the actual representation of data, which, in the second step, is used for completeness evaluation. The paper presents both theoretical background and experimental results performed on real-world RDF linked datasets.