Mining frequent patterns is an essential task in discovering hidden correlations in datasets. Although frequent patterns unveil valuable information, there are some challenges which limits their usability. First, the number of possible patterns is often very large which hinders their eff ective exploration. Second, patterns with many items are hard to read and the analyst may be unable to understand their meaning. In addition, the only available information about patterns is their support, a very coarse piece of information. In this paper, we are particularly interested in mining datasets that reflect usage patterns of users moving in space and time and for whom demographics attributes are available (age, occupation, etc). Such characteristics are typical of data collected from smart phones, whose analysis has critical business applications nowadays. We propose pattern exploration primitives, abstraction and refinement, that use hand-crafted taxonomies on time, space and user demographics. We show on two real datasets, Nokia and MovieLens, how the use of such taxonomies reduces the size of the pattern space and how demographics enable their semantic exploration. This work opens new perspectives in the semantic exploration of frequent patterns that reflect the behavior of di fferent user communities.