ModelRefs / Text Classification — AI Glossary

Text Classification — AI Glossary

Assigning predefined categories or labels to text spans or documents using ML models.

Overview

LLMs enable zero-shot and few-shot text classification without labeled training data: sentiment, topic, intent, toxicity, and routing classification via prompting. For high-volume production, fine-tuned encoder models (BERT, RoBERTa) are faster and cheaper. LLM-as-judge can generate training labels for smaller classifiers (distillation).

Reference details

Topic	applications
Last reviewed	2026-06-24

Example: Find the crossover before you pick the architecture

Zero-shot classification with a general model needs no labelled data and works the day you write the prompt — but you pay per call, and every call sends the instructions again. A fine-tuned encoder needs labelled examples up front and then classifies for a fraction of the cost, fast enough to run on modest hardware. The decision is a crossover, not a preference: below some daily volume the prompt wins because labelling is the expensive part; above it the encoder wins because inference is. The productive path is usually both — use the large model to label, then train the small one on those labels.

Commonly confused with

Classification assigns a label from a closed set; extraction pulls typed values out of text; generation writes new text. Framing a task as classification is what makes it measurable — precision, recall and a confusion matrix per class — which is why converting an open-ended judgement into a labelled decision is usually the first useful step.

When to use it

Reach for it when:

Routing, triage, moderation and intent detection, where the label set is known
Cold start with no labelled data, using a general model to bootstrap
High volume with stable classes, using a small fine-tuned model distilled from those labels

Reach for something else when:

Where classes overlap or shift constantly and the label set cannot be pinned down
Without a confusion matrix — aggregate accuracy hides collapse on the rare class that matters
For nuanced judgements that need a rationale, where a label alone is not actionable

Continue your research

Use these connected ModelRefs sections to compare alternatives, inspect implementation paths, and review the evidence and governance boundaries relevant to Text Classification — AI Glossary.

Frequently asked questions

What is Text Classification?