Labeling is a key aspect of any task management or ticketing platform.

At Dosu, we've invested heavily in auto-labeling because it's a time-saver for engineers, keeps projects organized, and enables Dosu to excel in everything else it does.

Approach#

Dosu's auto-labeling system works using the context of a project's current labels and past labeling activity to determine the relevant labels for newly created issues, pull requests, and tickets.

How It Works#

Dosu analyzes:

• Existing labels in your repository
• Historical patterns of how labels have been applied
• The text content of issues and pull requests

From this, it predicts the most relevant labels for each new item.
When you accept or reject a suggested label, Dosu learns from that correction to improve future accuracy.

Enabling Auto-Labeling#

1. Connect your GitHub repository to Dosu if not already connected.
2. Navigate to Settings > Channels in the Dosu dashboard.
3. If a channel doesn't already exist, create one by following the instructions in Key Concepts: Channel.
4. Go to Issues and toggle Auto-Label Issues.
5. Choose the labels you'd like to apply

Recommended Rollout#

1. Start with Preview Only
   Review Dosu’s recommendations without changing workflow.
2. Auto-Apply Low-Risk Labels
   Enable automatic application for safe, unambiguous labels like:
   • kind:question
   • area:docs
3. Expand Gradually
   Add more labels to the auto-apply list once accuracy is consistent.

Reviewing and Correcting Labels#

• To accept a suggestion: leave the label as-is.
• To reject a suggestion: remove the label from the issue or pull request.
• To add a missing label, apply it manually as you normally would.

Every action feeds back into Dosu’s learning model.

Best Practices#

• Keep labels clear and consistent
  Short, descriptive names improve prediction accuracy.
• Avoid duplicate or overlapping labels
  Merge similar labels to simplify classification.
• Regularly review suggestions
  Early corrections speed up the learning curve.

Usage In The Wild#

Dosu supports auto-labeling for GitHub issues and pull requests. See the repositories below for public examples of Dosu auto-labeling:

• https://github.com/apache/superset/issues
• https://github.com/apache/airflow/issues
• https://github.com/langchain-ai/langchain/pulls
• https://github.com/langgenius/dify/pulls

For more information on how to configure auto-labeling for GitHub, see the Auto-label Issues page.