Predictive Labeling
Predictive labeling helps improve prediction performance when preparing labeled data and reduces annotation time.
Introduction
Predictive labeling uses machine learning to generate label predictions based on a subset of manually labeled data. This feature reduces the time and effort required for manual labeling.
Once predictions are generated, they can be accepted or rejected. Predictive labeling is especially useful for large datasets, but predicted results should always be reviewed to prevent errors.
Key features
Predictive labeling provides the following benefits:
Efficiency: Reduces manual labeling effort by using model predictions.
Consistency: Helps maintain consistent labeling across large datasets.
Scalability: Supports labeling at scale for large volumes of data.
Quick start guide
Follow these steps to use predictive labeling in a row labeling project:
Enable Predictive labeling
In a row labeling project, click the gear icon in the extension panel on the right to open the Manage extensions dialog. Then, enable the Predictive labeling extension.
Configure input and output fields
Set the configuration in the extension:
Input column(s): Columns used as context for prediction.
Target field: Field where predicted labels will be stored.
Save configuration and start prediction
Click Save configuration to start the prediction.
If labeled data already exists, predictions will be generated immediately.
If not, label at least 5 examples per answer option first.
For example, if there are two answer options: POSITIVE and NEGATIVE , label at least 5 examples as POSITIVE and 5 examples as NEGATIVE.
Review predictions
Review generated predictions and accept or reject them. This iterative process helps improve labeling quality over time.
Last updated