Detect interesting patterns machine learning

I am starting work with a machine learning project for data risk management. Assume we have 1 million unstructured text documents, with some meta data like Data Owner for each text.(We can do any assumption)

One of the goal is to detect interesting patterns. But I do not really know what does this requirement mean. Could anybody tell me this? or give some related key words? (not ask for implementation) it's a quite open question. Hope interesting ideas!!