LLM-based Ontology Generation and Improvement

Overview

Ontologies, also known as data models, are one of the key concepts in Altair Graph Studio. Ontologies are the equivalent of a schema in an SQL database, but they provide a much deeper, semantic understanding of the data. They define the possible links between the classes in your knowledge graph. In a social network, for example, a data point could state that Martin is a friend of Joe, but an ontology would generically state that people have friends, people live in cities, cities are located in countries, etc. Ontologies give you the structure of the "world" you want to describe, be it a software bill of materials, an industrial factory with products and machines, or a financial institution with customers, accounts, and investment instruments. Thus, establishing an appropriate data ontology right is crucial. Graph Studio provides an LLM service can that offer suggestions to improve not only the labels of auto-generated ontology elements but also their descriptions, and URIs for classes and their properties.

As soon as you import data in a graphmart via any of the available options, Graph Studio automatically creates an initial version of the ontology based on names, shared columns, and other clues. This feature is a major step in helping you understand your data.

In some cases, however, the generated names may be incorrect. Data often include names with acronyms or numbers, rendering any user not directly involved in creating or maintaining the data unable to read or understand them. Graph Studio addresses this challenge by offering an LLM-based mechanism to automatically generate suggestions for better names and, very importantly, good descriptions. Thus, your ontology becomes truly semantic.

You may also obtain additional link (even data ingestion) recommendations based on the LLM suggestions. Graph Studio is exceptionally suited as a source of MCP tools for AI Agents. External agents can not only read data from the Graph but also automatically create and edit Graphmarts and ontologies based on data structure and their own context. This provides another step in automatically generating and optimizing data ontologies.

Enabling the Ontology Improvement Service

To enable LLM-assisted ontology improvement, the Ontology Improvement Service must be enabled and linked to a suitable LLM connection. This is done in the Admin Settings page.

Using the Ontology Improvement Service

To use the Ontology Improvement Service, open a Graphmart and select an ontology to be improved. Click on the three vertical dots that appear to the right of the Ontology panel and select Improve Ontology from the options that display.

When Improve Ontology is selected, the ontology data are sent to the LLM and processed, and the improvement suggestions are returned as results.

When the dropdown to the right of a label is expanded, a suggested label and description display.

By default, all items in the suggested ontology are selected. You can specify which of the available suggestions to apply to your ontology by selecting or deselecting an item. Click Save to update your ontology.