All terms
Data
Labeling
Assigning target tags or values to examples so a model can learn from them.
Definition
Labeling is the process of attaching a target tag or value to each example in a dataset, such as marking an image as a cat or rating a response as preferred. These labels supply the ground truth that supervised and preference-based training methods learn to predict. Labeling can be done by people, by models, or by a mix of both, and its accuracy directly affects model quality.