A system (50) for earning functional usage patterns of a user of the system (50) and for planning a best sequences of events suitable for a particular user in a particular context is disclosed. Information relating to user decisions, such as the context during which the decision was made and the actual user decision, is gathered by the DTV-agent (36) and delivered to the active avatar agent (37). The learning module (39) operates to identify all generalization patterns from a number of instances. Method (500) then determines which particular decision is most appropriate by comparing the generalization patterns with the current context for which a decision must be made. Clusters of such behavior patterns are formed with each cluster having the same number of matched attributes. A decision selecting process then selects a decision which is most appropriate to the current context from the behavior patterns.