Data Labeling

Data labeling, or annotation, is the work of attaching meaningful tags to raw data: marking objects in photos, assigning sentiment to reviews, transcribing audio or ranking model answers. These labels become the ground truth that supervised learning algorithms train on and are evaluated against.

Why it matters

Model quality is capped by label quality. Inconsistent, biased or wrong labels teach the model the same mistakes. Good projects write clear guidelines with examples, use several annotators per item, measure agreement between them, and audit samples. Labeling is often the largest cost in a machine learning project, so teams use active learning, model-assisted pre-labeling and review queues. Sensitive data may need access controls and compliance with privacy law.

Example

Text: "The dashboard is slow after login."
Label options: bug, feature_request, praise, question
Annotator 1: bug   Annotator 2: bug   Annotator 3: question
-> majority label: bug, agreement 67%, flag for guideline review

Common mistakes

• Vague guidelines that let annotators interpret classes differently.
• Never measuring inter-annotator agreement.
• Sending personal data to external labelers without safeguards.

Related

• Supervised learning
• Training data
• Classification
• AI bias

This page is general information written by our editorial team and updated when documentation changes. Servers and providers differ, so test changes safely first. See our editorial policy and disclaimer.