In this paper, we develop a method (WTRFS) that includes the user interaction in order to guide her through data mining results. This work aims to replace the descriptor declaration step used in interactive data mining. For this we exploit the hypothetical relation between experts' patterns of interest. We empirically demonstrate that WTRFS returns first the most relevant results for the user. Moreover, even if the users' interactions are not perfect, we observed that WTRFS behavior isn't altered.