In this paper we present a novel approach to location obfuscation for location privacy purposes. The motivating observation is that obfuscation strategies only grounded on geometric criteria can lead to privacy leaks. We thus propose to complement geometry-based approaches with obfuscation methods which account of the semantic context in which users are located. The unique contribution of the paper is the introduction of a comprehensive obfuscation framework consisting of a semantic-aware obfuscation model and a novel algorithm for the generation of obfuscated spaces of which we report an experimental evaluation.