Data classification in Microsoft 365 can be summarized as classifying documents and files according to their contents. Embedding labels in the documents and automated assessment of arbitrarily complex content are the key features of Microsoft's data classification. This functionality falls under the Microsoft purview umbrella and can be accessed through the compliance portal.

Microsoft purview uses sensitive info types and trainable classifiers to enable document labeling at scale. Sensitive info types are an extensive list of predefined information types that primarily use content formatting or keywords to determine the type of content. Using keywords, functions, or regular expressions, you can create your own sensitive info type. 

Trainable classifiers represent the next step in content assessment complexity using machine learning to assess content within OneDrive, Exchange, and Teams channels. Like sensitive info types, there are predefined classifiers in the form of machine learning models. These models can be adapted and re-trained using custom positive and negative sample data stored in SharePoint folders. As a machine learning model, a trainable classifier is not a trivial piece of technology, and it comes with some caveats. You will need an E5 subscription to access this feature. Microsoft estimates between 7 to 14 days to crawl your files with the predefined trainable classifiers. Training a new classifier is a time and resource-intensive exercise. You will need at least 50 samples of positive seed data for the initial model learning phase, then at least 200, although the more, the better, positive, negative, and ambiguous samples for the training phase.

Content explorer allows you to browse your content by assessed info type. Content explorer list viewer and Content explorer content viewer are the roles required to use content explorer. 

My name is Hallam Webber, and we've been looking at Data Classification with Microsoft 365.