One of the most urgent problems presented on the web scenario is the difficulty to capture effective user-oriented information. Search engines return tons of data which often are so general as well as useless; sometimes the user request is so specific that no matches occur or they are too poor to satisfy the user expectation. Often the same information is recoverable from different web pages and documents, but in an imprecise or contradictory manner, so returned results are vague or disappointing. In last years, on the light of the emergency of the semantics, the tendency is to model meta-data about the web resources through RDF-based approaches which assure appreciable machine-oriented understandability. Objective of this work is framed in a wider project in-progress for semantic-based information discovery. Herein, in particular, we focus on the classification of RDF documents in order to capture semantics inside web resources as a valuable alternative to deal with the traditional content-based view of the web information.