Submit a ticket My tickets
Welcome
Login  Sign up

Best Practices for Using Description Autogeneration

Empowering Stewards and Readers with AI-Powered Metadata Enrichment

To help you reduce manual work and enhance catalog completeness, DataGalaxy offers AI-powered description and summary autogeneration. This feature supports both Data Stewards and Data Readers, allowing them to benefit from intelligent suggestions while maintaining control over catalog content.

Overview

RoleAI's RoleUser's Role
Data StewardGenerate contextual suggestions based on object neighborhoodReview, accept, edit, or reject the suggestion
Data ReaderProvide informative autogenerated content for better understandingConsume suggestions and report inaccuracies

How Autogeneration Works

The AI analyzes metadata from the object and its neighborhood (parents, children, linked objects) to determine if there’s enough context to generate a meaningful, high-confidence description. If the required minimum context is met (at least one described object from two different link types), the system offers a suggestion.

We offer two options: automatic generation with manual approval and automatic generation by batch per workshop.

Best Practices for Data Stewards

1. Leverage the “Autodescription” Button

  • Look for the Autodescription button next to an attribute or object.

  • Click it to generate a description or summary suggestion if enough context is available.

2. Review Suggestions Thoroughly

  • AI-generated descriptions will be visible for review.

  • Evaluate accuracy, tone, and domain relevance before accepting.

3. Accept, Edit, or Reject Thoughtfully

  • You can:

    • ✅ Accept the suggestion as-is.

    • ✏️ Edit the suggestion after accepting it.

    • ❌ Reject if the description is not suitable.

4. Request Alternative Suggestions

  • Use the Autodescription menu to:

    • Regenerate alternative versions.

    • Report inaccurate suggestions to improve AI learning.

5. Update Your Catalog Regularly

  • When new links or metadata are added, previously unqualified objects may now qualify.

  • Recheck the Autodescription feature after catalog updates.

Best Practices for Data Readers

1. Encorage usage of AI Descriptions by stewards

  • Autogenerated content helps fill gaps in the catalog and supports quicker object comprehension.

2. Report Inaccurate Descriptions

  • If something looks off:

    • Use the report option in the Autodescription menu.

    • This feedback helps improve future suggestions for everyone.

Multilingual Support

For multilingual environments:

  • Once a suggestion is generated in one language, automatic translations are available.

  • You can still review and refine the translation if necessary.

⚠️ When No Suggestion Is Generated

If AI doesn’t provide a description:

  • It means the object doesn’t have enough context (i.e., insufficient linked, described neighbors).

  • Try enriching the catalog by:

    • Adding more metadata.

    • Describing related objects to unlock future autogeneration.

Thus, let's work together on your catalog completeness!

Do / Don't

DoDon't
Use autogeneration to save time and fill description gapsAccept suggestions without reviewing them
Regularly update object links and metadata for richer contextAssume AI can generate descriptions without context
Encourage Readers to report issuesIgnore inaccurate autogenerated descriptions


By collaborating with AI, both Stewards and Readers help create a smarter, more complete data catalog - with minimal effort and maximum value.

Best Practices for Automation

As soon as you checked automated generation with manual approval and proved that it matches your business case, do not hesitate to use automated generation by batch. It allows to generate missing descriptions and summaries for all eligible cases in a workspace.
The generation can be reverted if needed (if the new batch for this workspace was not started).

This option is available for administrators.


To check the automated generations - use the filters Description is not empty and Uprated On.

Note: we are working on the separate AI filter.



Did you find it helpful? Yes No

Send feedback
Sorry we couldn't be helpful. Help us improve this article with your feedback.