US Patent 11768916 Detection of sensitive database information

A computer system performs a scan to identify data items in a database that correspond to one or more of a plurality of specified output classes. The scan includes determining metadata for a portion of the database, and generating a data profile for a set of data items stored in the portion of the database. The generating includes performing a character-based analysis of the set of data items. The scan further includes identifying whether the set of data items corresponds to one of the plurality of specified output classes by utilizing a multi-class neural network classifier trained to perform the identifying using a plurality of features, including features extracted from the metadata and the data profile. The computing system identifies, based on outputs of the multi-class neural network classifier, a particular one of the plurality of specified output classes that corresponds to the set of data items.