A classifier maps information to labels rather than producing unrestricted prose. Its input might be a message, image or collection of measurements, and its output might identify the most likely category or give a score for each available category.
Classification is distinct from deciding what to do with the result. A separate policy can turn a label into an action, and that policy should account for uncertainty, mistakes and the consequences of acting on the wrong label.
ELI5
Think of a classifier as a sorter. It looks at an item and chooses which of the available groups it belongs in. The groups need clear meanings, and unfamiliar or confusing items can still be placed in the wrong group.
For example, an email service might sort messages into personal mail, receipts and unwanted mail. A receipt sent to the wrong folder shows why a quick choice is not necessarily a correct one.
Does a classifier have to generate text?
No. It can return a label, a number identifying a category, or a set of category scores instead of a written explanation.
Is classification the same as taking an action?
No. A classification describes which category fits the input. Another rule or system decides whether and how to act on that result.
