Classification vs ranking

Classification sorts something into one of a finite set of categories. Ranking orders a large set of items by a score. Reading a visitor's intent is classification into a handful of states; choosing items from a catalog is ranking. They are different problems with different right tools.

By Alex DeanPublished

Two different problems

Classification sorts an input into one of a fixed set of categories. Ranking takes a large set of items and orders them by how well each fits. Both can use a model, but the shape of the answer is different: classification returns one of a few labels, ranking returns an ordered list out of many.

A handful of answers versus many

Reading which of a few states a visitor is in is classification: a small, known set of answers. Choosing which items to show from a catalog of thousands or millions is ranking: that is what a recommendation model is built for, and it earns its keep there. Treating the intent question like a ranking problem is why it so often carries the weight of a full model it does not need.

Use the right tool for each

Keep the recommender for ranking the catalog. For intent, a typed question over the live state of the session sorts a visitor into one of your named states without a trained ranker. Signals supplies that live state and a model answers the typed question; the recommendation model goes on doing the ranking it is good at. The two sit beside each other rather than one replacing the other.

See it on your own traffic

Add the SDK, define one attribute, and see an attribute updating against your own traffic in the first session. The free tier runs on the real engine: no card, no sales call.