What is a classification model?

A classification model reads an incoming message and assigns it one or more predefined labels based on its content. It is the model type to use whenever you need to sort messages into categories you already know - for example detecting intent, tagging topics, or routing a ticket to the right team.

This article explains what a classification model does and the key concepts you will work with. To build one, see the "Set up your first classification model" article.

What it does

When a message arrives, the classification model compares its content against the labels you have defined and predicts the label (or labels) that best describe it. That prediction is then available to your automations and analytics - so you can, for example, route every message labeled "Billing question" to your finance inbox.

The model works from the meaning of the text, not from keyword rules, so it can recognize a label even when the wording is different from anything it has seen before.

Labels

Labels are the categories your model can choose from. You define them yourself, give each one a clear name and description, and the quality of those descriptions directly affects how accurately the model predicts.

Labels can be organized in a hierarchy. A top-level label can have child labels beneath it, which lets you model broad categories that break down into more specific ones (for example "Billing" with children "Refund request" and "Invoice question").

Single-label and multi-label

A classification model works in one of two modes:

• Single-label: the model picks exactly one label for each message. Use this when your categories are mutually exclusive.
• Multi-label: the model can assign zero, one, or several labels to the same message. Use this when a message can legitimately belong to more than one category.

The mode you choose also affects how accuracy is measured. See the "How classification accuracy is measured" article for details.

Predictions and feedback

Every time the model runs, it produces a prediction. Your team can accept a prediction or correct it, and those corrections become validated examples that improve the model over time. This feedback loop is how a classification model gets better at your specific business.

For more on what Tekst stores during this process, see What data does Tekst store and how does the learning process work?.

Related articles

• For an overview of all model types, see the "What are models on Tekst?" article.
• To build a model, see the "Set up your first classification model" article.
• To understand scoring, see the "How classification accuracy is measured" article.