Taxonomy Builder

23/06/2026

The Taxonomy Builder introduced in Graph Modeling 10.1 is a feature designed to simplify and accelerate the creation of knowledge graphs. It uses the capabilities of Large Language Models (LLMs) to provide an AI-assisted method for taxonomy creation and expansion.

This feature is specifically engineered to address the "cold start" problem often encountered in new knowledge graph projects by providing a rapid foundational knowledge model that users can build upon.

The Taxonomy Builder helps both taxonomists and knowledge-domain experts build and grow taxonomies quicker by automatically suggesting new concepts:

For professional taxonomists, the Taxonomy Builder functions as a productivity tool and a collaborative partner. It supports the manual process by providing AI-powered suggestions for new concepts, labels, definitions, and hierarchical relationships, allowing them to focus on quality assurance and governance.
For users with deep domain knowledge but lacking technical expertise, the Taxonomy Builder enables knowledge modeling. It simplifies this process by enabling non-taxonomists (including internal staff, partners, and customers) to create and expand taxonomies for pilot projects without needing deep technical understanding of semantic technologies.

The Taxonomy Builder provides precise control over the taxonomy creation and refinement process by allowing users to:

Control the generation of concepts by formulating an individual prompt for the LLM
Specify a number of parameters to further control the output of the generation process, such as:
- Which concepts in the existing taxonomy should be extended
- Which terms should be included in and excluded from the taxonomy
- How deep the generated taxonomy should be as well as how many child nodes should be generated per concept
- The temperature of the LLM, i.e. the randomness and creativity of the generated output. Low temperature corresponds to a more deterministic output, whereas high temperature leads to more diverse and sometimes surprising generated content.
Generate additional concept attributes such as alternative labels and definitions, thereby including features already available in the Extend your Taxonomy tab.
As of Graph Modeling version 10.3, the Taxonomy Builder allows user to localize both existing and generated labels and definitions into various languages. This feature is critical for expanding taxonomies beyond the default project language.

To access the Taxonomy Builder, navigate to the Thesaurus Management view of your Graph Modeling project and either:

right-click any node in your taxonomy tree (1) and then hover over the Build Your Taxonomy (2) context menu item to expand and select one of the two sub-menu items: Build Hierarchy (3) or (Add Details) (4).
Tip
The Taxonomy Builder supports multi-select and bulk operations, allowing you to expand multiple nodes at once.
click the Generate Taxonomy button (5) on the Details View of your project node. This will open the Taxonomy Builder with the Build Hierarchy tab open by default.

Note

If you cannot access this feature, visit our help page Configuring External LLM Services for more information on how to set up and configure LLM-based features such as the Taxonomy Builder.

The Taxonomy Builder uses LLMs to automate the creation of structured taxonomies which currently takes place across two stages:

Build Hierarchy: An LLM-based approach to generate the foundational hierarchy skeleton, URIs, and preferred labels. Generated suggestions are based on an optional prompt and other generation parameters.
Add Details: LLM-based generation of alternative labels and definitions in the default language to ensure concept clarity and search precision. This stage also allows the localization of preferred labels, alternative labels, and definitions into other languages.

These workflow steps can be used together to either build a taxonomy from scratch or expand an existing one. Alternatively, you can use them independently to suit your specific taxonomy-building and fine-tuning needs.

The subsequent pages of this chapter provide an in-depth look at how each workflow step works.

Note

Advanced users can adjust parameters before starting the workflow to achieve more granular control.

The following advanced Taxonomy Builder settings can be accessed via the gear icon (1) next to the Generate button:

Source parameters (1): Provides an overview of existing concept and concept scheme descriptions in the selected taxonomy. Detailed descriptions already present in your taxonomy lead to higher quality results during generation.
Generation parameters (2): Here you can find the following configuration options:
- Narrower Concepts to exclude: When a node is selected on the first page of the Taxonomy Builder, its narrower concepts are automatically included as context. This setting allows you to select specific concepts to be bypassed during the generation process. While these concepts remain part of your taxonomy, the LLM will not take them into account or generate any further sub-concepts from them.
- Terms to include: Provide a list of strings or terms. This guides the LLM by offering examples of the type of candidate concepts you are looking for.
- Terms to exclude: Provide examples of terms or topics that should be avoided during generation. These serve as negative examples to help the LLM understand the boundaries of your domain; the system will aim to exclude both these specific terms and any conceptually similar ideas.
Additional concept attributes (3): Here you can enable or disable Generate Alternative Labels and Generate Definitions. This functionality is also described on our help page on the Extend your Taxonomy tab.
Advanced parameters (4): Here you can enter a decimal number from 0-1 to define the Temperature of your generated results. Temperature is a parameter that controls the randomness of a large language model's output, with lower values making the output more predictable and higher values making it more random. You can also set the maximum number of tokens in the LLM response.
Note
The response will be truncated once it reaches the token limit and will probably be a malformed JSON. If you want more concepts and depth you can increase the number of tokens.

Once the desired parameters have been set, select Set (5) to apply. Click Cancel (6) to exit out of the dialog without saving.

The next section explains in detail how to build the hierarchy of your generated taxonomy using the Taxonomy Builder.

Upcoming Developments

Future releases of the Taxonomy Builder will expand beyond the initial Generative AI approach to incorporate a document-based approach:

Expansion from Documents: The Taxonomy Builder will gain the ability to expand an existing taxonomy based on user-provided documents. This grounded, document-based expansion method increases the taxonomy's quality and relevance to the source material. Relevant terms are automatically extracted from the documents to form candidate concepts and then incorporated into a hierarchical structure that is aligned with the original taxonomy.
Adding Attributes and Relations: Using both LLM and corpus-based methods, the Taxonomy Builder will generate and propose SKOS relations, as well as ontology-based attributes and relations. This capability transforms simple concept lists into robust knowledge models supporting the use of sophisticated applications, inference engines, and deeper analytics.

In this section: