An application can provide a question, supporting information and a fixed set of choices. The model scores the choices so the application can use the result in a larger workflow.
The available choices need to represent the real decision. For example, when sorting messages, an application may need an uncertain result so a person can review unclear cases.
ELI5
For example, imagine a help desk with bins labelled Billing, Login and Other. A decision model reads a message and gives each bin a score so the help desk knows where to send it.
If the message says both that a bill is wrong and that the sender cannot sign in, the scores may be close. The help desk can ask someone to read the message before choosing a bin.
How does an app use the result?
It can use the chosen answer or its score to decide what to do next, such as replying, waiting or asking a person to check.
Does choosing from a list prevent mistakes?
The list constrains the possible answers. The model can still choose the wrong answer, so check its results on representative examples.
