Data Annotation
Definition
Data annotation is the process of adding labels, categories or structured judgments to raw data (text, images, audio or video) so that it can be used to train or evaluate a machine learning model.
Models learn from examples. Annotation turns raw data into those examples: marking objects in an image, classifying a document, transcribing audio, or judging whether an answer is correct. The labels become the ground truth the model is trained and measured against.
Annotation quality sets the ceiling on model quality. For subjective or technical data, the label depends on the annotator's judgment, so who annotates matters as much as how many annotations there are. A second rater on a sample, clear guidelines and disagreement review are standard quality controls.
Annotation can be as simple as a yes/no tag or as involved as a structured expert assessment with written reasoning, depending on what the model needs to learn.
How Pathwize relates
Pathwize provides annotation by subject-matter experts, with a qualification test on your own data, a sampled second rater, disagreement review, and a record of who labelled each item and how long it took.
Frequently asked
Adding labels or judgments to raw data so a model can learn from it or be measured against it.
Labels are the ground truth; if they are wrong, the model learns and is scored against wrong answers.
They mean the same thing. Labeling usually refers to simple tags; annotation often implies richer, structured judgments.
Related terms