Link Search Menu Expand Document
Start for Free

Ontology Maintenance & Data Mapping

This guide explains how to use Voicebox for Designer to maintain ontologies, connect and map data sources, and validate your knowledge graph.

Page Contents
  1. Overview
    1. Key Functionality
  2. Example Workflow
  3. Requirements
  4. Voicebox evaluations
    1. Model Summary
    2. Questions

Overview

Voicebox for Designer is a specialized Agent that guides users through the various steps of creating and curating a knowledge graph: creating an ontology, mapping data sources, and aligning model and mappings to your specific user case. This allows SMEs to build sophisticated knowledge graphs without requiring deep ontology expertise. Designer provides an intuitive, low-code interface that helps knowledge workers define end-user questions, connect data sources, and publish ontologies to support their knowledge graph use cases.

This feature is designed to address common data modeling and mapping challenges:

  • Ontology creation has historically been a specialized skill, leading to project delays or the need to hire additional resources.
  • Enterprises often need domain-specific context that generic AI models cannot provide.
  • Achieve faster time to value results from your knowledge graph.
  • Avoid time-intensive training or other large upfront investments to get started as Voicebox does not require these.

Key Functionality

  • Automated Ontology Generation: Automatically map data sources to ontology concepts so your knowledge graph is customized for your domain via your data and your questions.

  • Human-in-the-Loop Oversight: Review mappings, resolve conflicts, and validate provenance.

  • Ease of Use: Low-code/no-code interface that minimizes manual setup.

Example Workflow

A typical ontology creation workflow:

  1. Define the scope of your project
  2. Define a project title
  3. Provide a project description. Write this from the perspective of what your project will be used to accomplish. Think of this as a user story or a problem statement.
  4. Review and write questions which will be commonly asked of your knowledge graph
  5. Review and select data sources to support your project
  6. Let Voicebox process your inputs. Voicebox may take a few minutes to analyze your questions and data sources.
  7. Review the suggested ontology and mapping from Voicebox
  8. Iterate by adding data model concepts, connecting additional sources, or refining your project questions until Voicebox shows high confidence in answering them.
  9. Publish to your Stardog Endpoint.
  10. Interact with your knowledge graph using Voicebox, a conversational AI agent available through Stardog interfaces and API.

Stardog Academy has a video demonistrating this workflow.

Requirements

Voicebox for Designer requires a Voicebox enabled endpoint.

Voicebox evaluations

Model Summary

  • Voicebox chat can be opened by clicking the Voicebox blurb at the bottom left of the canvas.
  • The model summary is available as a collapsible section within the Voicebox chat.
  • The model summary provides a detailed explanation of your project and lists the recommended actions you can take to improve your knowledge graph.
  • The recommended actions can include:
    • Unanswered questions. Voicebox may not be able to answer these questions with high confidence. Review the details to learn more.
    • Unused resources. To include all data in your knowledge graph, ensure each resource is mapped to a data model concept.
    • Unused data model concepts. Your model includes concepts that are not linked to others. Connect them to build a complete data model.
    • More questions recommended. Your project currently has too few questions to evaluate. Add questions to illustrate how your knowledge graph will be used and to improve model and mapping accuracy.
    • More resources recommended. Your project has no resources, and therefore no data has been mapped. Add resources to begin populating your knowledge graph.

Voicebox-blurb

Voicebox-model-summary

Questions

  • Each question will be scored against the data model and mapped data in your projects. A low score is likely not answerable by Voicebox. A high score is likely answerable.
  • Questions are scored based on:
    • Clarity, The question is ambiguous.
    • If concepts required to answer the question are captured in the data model.
    • If concepts required to answer the question have mapped data in the project.
      • Please note, question evaluation is limited to the scope of the project. Therefore, question evaluation does not take into account data defined for model concepts outside of the project.
  • Each question can be expanded to show more details about the score. To see a summary explaining the score, click More details

Voicebox-question_eval