We use a taxonomy management tool called PoolParty for Content classification. It has an “extractor” module that processes text we give it and matches strings in that text with concepts in our Taxonomy. For autotagging, we give it the title and body text of articles on the site, and it processes the text and returns the 1 or 2 products that score most highly. Overall, it has a very high level of success (our median agreement between a taxonomist tagging articles and PoolParty tagging articles is 90%), but it isn’t perfect and does make mistakes. These mistakes happen mostly with very generically named products like “Azure Files” which get applied when they’re not relevant, but an article mentions “Azure” and “files” repeatedly. We’re rolling this out for site search first, where it isn’t supplanting any existing features. Next year, I expect the platform to invest in more features that will let us fine tune the tagging, integrate it with author review, and use it in more user-facing contexts.