Why Ontologies Matter for AI Success
14 entriesOntologies define what the things in your data mean: their classes, relationships and constraints, in a form that people and machines can both use. Much writing on AI and ontologies focuses on retrieval-augmented generation, where knowledge graphs supply context. A 26 October 2026 LinkedIn article by Andrea Splendiani, the first part of a series titled "Ontologies for AI in practice", looks at a different question: what happens when the data itself carries more semantics? In a small, deliberately informal experiment on a fictional family dataset, the author asked an LLM to count a person's ancestors and added semantics step by step: data only, cleansed data, plain-text definitions, an RDF knowledge graph, a curated knowledge graph and finally an ontology. The author reports that data alone confused the model, light cleansing changed little, simple definitions gave a large improvement, RDF exposed hidden assumptions, curation improved trust and consistency, and an ontology gave the model a further edge. The author stresses that the experiment is not representative or unbiased. Descriptions below are paraphrased, are not rankings, and release facts come from GitHub and PyPI on 1 November 2026.
Semantic Web Standards
11 entriesW3C specifications that underpin ontologies and knowledge graphs.
Ontology and Semantic Tools
8 entriesEditors, libraries and engines for building and validating ontologies.
Domain Ontologies and Examples
4 entriesEstablished ontologies that show how shared semantics are used in practice.
Frequently Asked Questions
Do ontologies actually help LLMs?
In one small experiment reported on LinkedIn by Andrea Splendiani, adding semantics improved an LLM's answers: plain-text definitions gave a substantial improvement, a curated knowledge graph improved trust and consistency, and adding an ontology made a further difference. The author notes the experiment is far from representative or unbiased, so treat it as an illustration, not proof.
What did the experiment test?
It asked an LLM to count a person's ancestors in a small fictional dataset that deliberately left some ambiguity, then re-asked after adding cleansed data, definitions, an RDF knowledge graph, a curated knowledge graph and an ontology. It was run in a single session, so later steps could see earlier ones, and the author reports that execution time and approach varied between runs.
What is the difference between an ontology, a knowledge graph and a semantic layer?
An ontology defines concepts, relationships and constraints. A knowledge graph holds the data as entities and relationships, often described by an ontology. A semantic layer defines business terms and metrics for analytics. They are complementary. See the Ontologies, Knowledge Graphs and Semantic Layers pages.
Do I need RDF and OWL to give an LLM semantics?
Not always. In the experiment, plain-text definitions already gave the biggest improvement for the effort, while RDF and an ontology added precision and exposed hidden assumptions. Start with clear definitions, then formalize where consistency and reasoning matter.
Where should I start?
Write definitions for your key data elements, model the core concepts in an ontology tool such as Protégé, validate data with SHACL, and store it in a triple store or alongside your catalog. See the Ontologies and RDF Graph Databases pages.
What is Apache Ossie?
An incubating Apache project, formerly the Open Semantic Interchange, aiming to standardize how semantic models are exchanged across analytics, AI and BI platforms. Sources describe its ontology work as planned, so check the project site for the current status.
How current is this page?
Links and release facts were checked on 1 November 2026 from GitHub and PyPI. The LinkedIn article may ask visitors to sign in. Protégé and RDFLib have had no release since early 2026, as their cards note. Descriptions follow each source's own pages and are not rankings.