Ontology Pipeline
A six-stage practitioner method for building an organization's formal knowledge layer, in order: controlled vocabulary, metadata schema, taxonomy, thesaurus, ontology, knowledge graph. Jessica Talisman set it out in Ontology Pipeline: A Framework for Knowledge Engineering (Technics Publications, published 2026-09-09). Where this spoke's synthesis describes the rdf → RDFS/OWL → SHACL/SPARQL stack as layers of expressive power, the book reads the same stack from the builder's side: the order to build it in, and what it costs to keep true over time.
The six stages
- Controlled vocabulary — corpus analysis, definitions, and synonym control, grounded in the library-science standard ANSI/NISO Z39.19, with a section on using SKOS for AI.
- Metadata schema — structure and context, contrasting library-science practice with enterprise metadata-application-profile practice.
- Taxonomy — a navigable concept hierarchy, the "Is-a test" for a valid parent-child edge, and converting a spreadsheet taxonomy into SKOS.
- Thesaurus — broader/narrower/related term relationships and lightweight inference. The book argues SKOS is sufficient at this stage rather than a stepping stone to something heavier.
- Ontology — formal classes, properties, and rules in RDF, RDFS, and OWL; a "three-box architecture" pattern and competency questions as the acceptance test for a domain model.
- Knowledge graph — the assembled result, queried across systems. Enterprise, domain, and linked-data patterns, plus staffing, ongoing funding, change management, and what the book calls semantic debt.
Talisman frames the stages as composable and decomposable: a team can stop at thesaurus, or restart from taxonomy, instead of treating "build a knowledge graph" as one all-or-nothing project.
The argument for it
Talisman's case is that large language models, retrieval-augmented generation, and entity resolution succeed or fail on the quality of the data underneath them, and that library and information science already has a method for building controlled, governed description of what an organization knows. She frames this as enterprise data management expanding into enterprise knowledge management, with AI as the reason it's happening now rather than the reason it's needed.
Provenance and tier
T3. Talisman is presenting her own trademarked framework, which she also sells through contextually-llc consulting and knowledge-graph-academy training — a direct commercial interest in readers adopting it. The endorsement quoted on the book's sales pages is from Malcolm Hawker, Chief Data Officer at Profisee (a data-governance vendor), which is interested rather than independent. What sets it above a vendor blog post: 461 pages through a specialist publisher (technics-publications, not self-published), and named enterprise deployments — an RDF-based knowledge graph at Adobe, work at the Department of Justice — rather than unsupported claims. Kindle $39.95, print $39.95, PDF $34.99.
Fetch note. The Amazon link supplied with this ingest (.../dp/B0D4M3K5ZX) 500-errored on two
WebFetch attempts and does not match any listing a search could find. The live Kindle edition is
ASIN B0HJ8QWJ5V (paperback: B0HJ9223FH), recovered by scraping that ASIN with firecrawl after
Amazon's own page kept failing WebFetch. url: on this page points to the publisher instead of
Amazon, since Technics Publications' page carries the fuller table of contents and doesn't depend on
a format-specific ASIN.
Connections
- Reads rdf → RDFS/OWL → SHACL/SPARQL as a build sequence, not just a layered stack.
- Uses SKOS at the taxonomy and thesaurus stages — a vocabulary this spoke names as a coverage gap (see synthesis → Growth edges, item 7).
- By jessica-talisman, published by technics-publications.
Related
knowledge-representation · rdf · schema-org · protege · ontology-dev-tools