Entity Classification Service Overview
This engine experience introduces a reworked classification hierarchy, seeded detector types, and simplified likelihood buckets so users can better address common use cases and readily understand the classification results presented in the application. System classifiers will address a wide range of common classification types. In contrast, custom classifier creation is streamlined with two detector options, plus the ability to copy the system classifier template and customize as needed.
Entity Classification Service Hierarchy
Classification Profiles
A profile represents a group of associated or related classifiers based on a particular need, law, or framework, threshold level, or targeted data sample set.
For more information on creating classification profiles, see Creating & Managing Classification Profiles.
Classifiers
A classifier is a type of algorithm or model used in machine learning to categorize or classify data into different classes or categories. The goal of a classifier is to predict the category to which a new, unseen data point belongs based on its features. Once a data source is scanned, you can approve, flag for review, reject, or delete the recommended classification to accurately label your data assets and help manage your data landscape.
For more information on classifiers, see the following:
Detector Types
Detectors offer a few approaches to classifying a data entity. Patterns consist of regular expressions to validate data based on character set matching, while Dictionary types include a short list of possible keyword values to match against. You can also start from scratch by choosing the detector template of an existing classifier. A classifier can have multiple detectors to improve or narrow search conditions and classification requirements.
Detector types are read-only for system classifiers and cannot be edited. You can clone a system classifier's default template to customize it beyond the standard detection rules.
For more information, see Understanding Classification Likelihood Results.
Detection Rules
The automation logic behind Detectors, detector rules are run automatically in sequential order when a data entity sample matches the supplied detector type values. The rules will check multiple contextual clues to refine the likelihood that the PII was matched by a base detector (e.g., the column name if PII occurs in a table or leading keywords in file names).
Detector rules are read only for system classifiers and cannot be edited.
Reviewing Terms
The review experience for terms pending review or flagged for review has improved with the new service. Terms now roll up into the individual label sections, which provide progress details, additional metadata, and options for quick or in-depth review. For more information, see Reviewing Recommended Terms.