Classification

The task of assigning an input to one of several categories.

Which desk should receive this request?

A reader writes “The online catalogue cannot renew my book.” An archive wants to route the message to a service desk. The team must define what its categories mean before “correct route” can be scored.

Single-label routing

If exactly one desk owns the initial response, the model might choose circulation from a list that includes circulation, technical support and permissions. The label guide needs a rule for mixed requests.

Multi-label tagging

If the archive is describing topics, both circulation and online systems may apply. Forcing one label would discard useful information and change the task.

What does the predicted category mean?

A class is a category defined for this task. A score may express confidence or a model-specific decision value. Before treating it as a probability, check how the system produced it and whether its estimates match observed outcomes.

An envelope lies in front of three filing trays.
Routing a message requires a policy for choosing among the available categories.

How would you review an ambiguous case?

Read the request alongside the routing policy and record the desk that actually resolved it. A correct topic tag and a successful handoff can differ. Review confusion between desks as well as the overall fraction routed correctly.

Follow the idea

Supervised Learning · See how examples connect inputs to categories.

Binary Classification · Examine a task with exactly two classes.

Examples in the AI guides

More chapter examples (4)

Source · Google’s classification guide

Return to Classification in the A–Z glossary