Ability to automatically tag true content "aboutness"

The capability to predict subject tags (concepts) that a piece of content is actually "about" rather than purely predicting subject tags that have been mentioned.
What is "aboutness" in tagging terminology?
  • "Aboutness" refers to the true topic of a piece of media content (text / image / video /.. etc), and not just what it "mentions" or "depicts".
  • When using classic NLP, generic AI auto-tagging systems, or NER systems on text, you tend to get tag predictions that include all the things that are mentioned.
  • As a meta example, this FAQ mentions "NLP", "FAQ", "Media", "Mentions", "text", "image", "video" and more, but what it is really about is "aboutness" (and also "tagging")
  • A true "aboutness" auto-tagging system learns not what is mentioned in the text or seen in imagery, but what a subject matter expert determines that the content is about.
What's the difference between "about" and "mentions"?

In content tagging terminology:

  • "About" is used to describe the subject(s) (or "concepts") that a piece of content is truly about.
  • "Mentions" is used to describe the concepts that are mentioned, and this tends to be more verbose,
  • "About" tends to be more useful in driving discovery, engagement, and user journeys that carry meaning, whereas "Mentions" is useful for very detailed archive indexing.