Servers for a content provider are organized into three architectural layers that operate on three distinct time scales, including an offline layer, an intermediate layer, and a real-time layer. The servers (e.g., on the intermediate layer) obtain a periodic service attribute corresponding to a terminal identifier of a user terminal and a real-time service attribute corresponding to the terminal identifier. The servers (e.g., on the intermediate layer) select recommended content in a content data set of the content provider according to the periodic service attribute and the real-time service attribute, wherein selection is performed in the intermediate layer with a time scale between the update cycle of the offline layer and the response time requirement of the real-time layer, and corresponding selection processes are performed on an on-demand basis for a plurality of terminal identifiers corresponding to user terminals that are currently actively engaged in interaction with the real-time layer.