The discovery of interesting patterns is a challenging task in data mining. On one hand, approaches have been proposed to automatically learn user-specific ranking functions over patterns. These approaches are often accurate, but very time consuming. On the other hand, many interest measures are used to evaluate the interest of patterns while being as close as possible to a user interest. In this paper, we express the learning of a pattern ranking function as a multicriteria optimization problem. The proposed approach aggregates all measures into a single weighted linear function, where coefficients are computed via the Analytic Hierarchy Process (AHP). Experiments conducted on many datasets show that our approach drastically reduces the execution time, while ensuring a high quality ranking compared to existing approaches.