Metadata

Metadata is data about data. Data, by itself, doesn't mean anything. One of the primary ways we ascribe meaning to data is by associating it with "meta- data" that specifically tells us something about the instance in question. Typically the metadata is stored in data dictionaries or classes or somewhere where it need not be repeated for every instance of the data. We might store dozens of types of data about a datum, including the following:

• Physical representation information (stored as an "int" or a "float")

• Precision information (stored as a "float" or "double")

• Validation (validation that this item has already passed, such as "must be >50")

• Source (where the data came from, including program and user)

• Owner (who owns the data; e.g., who has which rights in its duplication)

• Temporal data (when the data was captured and posted)

• Display information (default display characteristic of this item; two decimal places)

• Name of the term (what the datum is called)

• Metadata type (attribute, entity, relationship)

• Semantic type (measurement, category, etc.)

Traditionally, metadata is static; we set it up in the data dictionaries and programs, and each instance takes on the metadata it had when it was created. But a metadata-inspired design recognizes that certain types of routine changes require changes to the static parts of the metadata (database definition, classes, programs, etc.). By shifting some of the metadata from the static realm to the dynamic realm, the design can be made much more flexible and reliable at the same time. Chapter 6 provides a detailed description and case study.