Manage > Vector Index

This section from the Manage Tab is only shown for asset collections that have the TopBraid AI Service collection included (see Enabling the Vector Index for an Asset Collection). It is the building block for AI features such as AI Linking and Natural Language to SPARQL.

Clicking Configure NL to SPARQL opens a dialog where you choose how to set up Natural Language to SPARQL for this asset collection. Both options replace manually configuring the Vector Index for the ontology and the asset collection, and creating a Natural Language to SPARQL configuration, as three separate steps.

Choose a configuration mode

Two configuration modes are available:

  • Automatic: The system analyses the ontology and asset collection and selects classes and properties based on relevance. This is the fastest option and works well for most collections. You can refine the generated configuration later using manual configuration.

  • Manual: A step-by-step wizard lets you choose classes, properties, and search parameters yourself. Use this when you need full control over the configuration.

Automatic configuration

After choosing Automatic, select the ontology to use for generating SPARQL queries from the list of ontologies imported by the asset collection, then click Start Configuration.

The following Advanced settings are available:

  • Class threshold (default: 100): The minimum number of instances of a class that must exist in the asset collection for that class to be included in the Vector Index configuration.

  • Property threshold (default: 10): The minimum number of usages of a property that must exist in the asset collection for that property to be included in the Vector Index configuration.

  • Clear the existing configuration (default: on): When enabled, any existing Natural Language to SPARQL and Vector Index configurations on both the ontology and the asset collection are removed before creating new ones.

  • Push existing data (default: on): When enabled, the Vector Index for both the ontology and the asset collection is built immediately after the configuration is created, running in the background. Turn this off if you want to review the generated configuration before triggering indexing.

Manual configuration

After choosing Manual, a three-step wizard walks you through the setup.

Step 1 - Select ontology

Choose the ontology to use for generating SPARQL queries, from the list of ontologies imported by the asset collection. The step shows whether the ontology and the asset collection are already indexed, so you know upfront whether continuing will trigger new indexing work. You cannot move to the next step until an ontology has been selected.

Step 2 - Configure

  • Classes to index: Use the + button to select classes from the ontology. Each selected class is shown as a chip with its instance count in the asset collection, and can be removed again.

  • Properties: Configure the properties that are shared across all the selected classes. Likely candidates, such as label properties, are suggested automatically; use the + button to add others manually. At least one property must be selected, and at least one selected property must be usable as a keyword for the keyword and hybrid search methods to work well.

  • Search parameters: Choose the search method - Keyword (exact and fuzzy text matching), Vector (semantic similarity search), or Hybrid (a combination of both, weighted by the alpha setting) - and, optionally, a threshold to filter out low-confidence results.

You cannot move to the next step until at least one class and one property are selected.

Step 3 - Review and start

Review the ontology, the selected classes and properties, and the search settings, then check the list of indexing tasks that will run. Classes and ontologies that are already indexed are marked as such; the rest are indexed automatically once you click through. If both the ontology and the asset collection are already indexed, Natural Language to SPARQL becomes available immediately, with no additional indexing.

This step also has a Clear configuration and vector index option, which permanently removes the Natural Language to SPARQL configuration and the vector indexes for both the ontology and the asset collection. This cannot be undone, and setting up Natural Language to SPARQL again afterwards requires going through the wizard from the beginning. This option is only available in manual configuration.

Indexing status

Once configuration is started, indexing runs in the background when push is enabled (automatic configuration default, or when indexing is required in the manual wizard).

  • The Vector Index section on the Manage tab shows per-task progress while indexing is running. When indexing is complete, an Edit configuration button lets you reopen the configuration flow.

  • An Indexing… badge in the collection header appears while indexing is in progress. Click it to open a dialog with per-task details. An Index error badge appears if indexing fails.

  • A notification is shown when indexing finishes or fails.

See Also

Further Reading on TQ Data Foundation