The approaches assume a set of training data, which provides, for example, the number of pages viewed by users in each cluster???corresponding to the number of impressions in each cluster???as well as the number and identification of ads clicked on in each cluster, and other information, such as demographical information about the users.