A method of establishing at least one variable of a new view of data, is disclosed where the variable is obtained from at least one data source (150-153). The method identifies at least one display type (1202, FIG. 15, 17 18), each display type having at least one corresponding template of data to be used by the new view of data. The template includes a plurality of data component slots (1250-1254) related according to the semantics of the identified display type. One or more data components to be included in the view of data are then identified, each identified data component being associated (1222) with a slot in the template. A list of recommendations is then formed for one or more unspecified data components of the new view of data, the recommendations being based on the semantics of the display type and the identified data components. Finally at least one data component from the list of recommendations is identified as a variable for said new view of data. Also disclosed is a method of learning unknown relationships among data components across a plurality of heterogeneous data sources, and a method for recommending relevant data components for view creation across a plurality of heterogeneous data sources.