Seeking relationships in multi-dimensional datasets is a common task, but can quickly become tedious due to the heterogeneity and increasing size of the data. Its visualization can be approached in a variety of ways: with projection techniques, overview techniques, and tabular techniques. However, while the interactive selection of a data subset during exploration is most easily done with tabular visualizations, finding relationships and patterns is not. Also, with overview techniques the number of attribute combinations quickly outgrows reasonable dimensions. In this thesis, a data-driven touring process for Visual Analytics (VA) tools is presented that guides users in discovering relationships for a data subset of their interest. Based on the users selection, attributes that show some kind of similarity are presented. The selection can be done on attribute and item level. While a selected attribute is compared to all other attributes in the dataset, item sets are compared to the individual categories of attributes. This comparison can be based on a number of similarity measures. To cope with heterogeneity of data types, numerical attributes are discretized to achieve maximum similarity. In hierarchical attributes, the most similar subtree is sought. The touring process is also independent of the data domain and its visualization. This independence is demonstrated by the use of three different datasets and the integration of the touring process into two VA systems.