We already have some understanding of the relationship between different metadata components. For example, there is a well-defined concept of a work contact, made up of particular metadata elements like phone number and email, and associated with a particular person. While we can derive much of this information using a data mining approach, it may prove effective to specify certain well-defined relations as a basis for further metadata pattern detection.
Part of the analytics research to be performed in CEDAR will evaluate whether user-defined structural relations, such as this one, actually improve the ability of the analytics framework to predict metadata values. If the framework’s predictions are as accurate without manual intervention, then we can save the trouble of deciding which relationships to define for the system, and simply let the system figure out the associations for itself.