The personalization and device diagnostic obviously are not the only uses of unstructured data. The preceding case is a good example as we started from structured record and quickly converged on the need to construct an unstructured data structure to simplify the analysis.

In fact, there are many more unstructured data than there are structured; it is just the convenience of having the flat structure for the traditional statistical analysis that makes us to present the data as a set of records. Text, images, and music are the examples of semi-structured data.

One example of non-structured data is denormalized data. Traditionally the record data are normalized mostly for performance reasons as the RDBMSs have been optimized to work with structured data. This leads to foreign key and lookup tables, but these are very hard to maintain if the dimensions change. Denormalized data does not have this problem as the lookup table can be stored with each record—it is just an additional table object associated with a row, but may be less storage-efficient.