From spreadsheets to a knowledge graph¶

An ontology says what kinds of things there are and how they relate. A knowledge graph is the things themselves - this pizzeria, that baker, the pizza baked at 18:05 - described with the ontology's terms. OntoBoard builds one from spreadsheets, inside Protégé.

Keep the data apart from the ontology¶

Put the graph in a file of its own that imports the ontology, rather than adding individuals to the ontology. The ontology stays reusable by anyone, the data can change every day without a new ontology release, and the graph says which version of the ontology it goes with.

@prefix ex:  <https://example.org/pizzeria/data/> .
@prefix owl: <http://www.w3.org/2002/07/owl#> .

<https://example.org/pizzeria/data.owl> a owl:Ontology ;
    owl:imports <https://example.org/pizza/1.0.0> .     # the ontology release it is made for

Beside it, a catalog-v001.xml that maps that address to the release file on disk, so Protégé opens the graph without the network:

<uri name="https://example.org/pizza/1.0.0" uri="../releases/1.0.0/pizza.owl"/>

1. One sheet per kind of thing¶

Each sheet is a ROBOT template: the first row is your own headings, the second says what each column is, and every row after that is one thing.

ID	Type	Name	Located in
`ID`	`TYPE`	`AL rdfs:label@en`	`I ex:locatedIn`
`ex:P_0000101`	pizzeria	Pizzeria Da Luigi	Karlsruhe
	pizzeria	Pizzeria Napoli Centrale	Naples

TYPE names the class each row is an individual of - by its label, as the ontology writes it.
I <property> relates the row to another individual - by its label or by its ID. Karlsruhe above is a row of the cities sheet: a sheet can name the rows of every other sheet in the same folder, in any order, and two sheets may name each other. ROBOT's command line and ODK read sheets one after another and find a label only in a sheet already read, so there a sheet can name by label only the sheets given before it - Check sheets... says in which order, and which sheets have none. A row named by its ID, without quotes, works in any order: ROBOT expands the ID itself.
The prefix of the IDs has to be declared in the graph file. ex:P_0000101 means something only where ex: is declared: in Protégé, the Active ontology tab → Ontology Prefixes, then save; in Turtle, @prefix ex: <https://example.org/pizzeria/data/> .. Undeclared, every row of the sheet fails with ROBOT's could not interpret 'ex:P_0000101'. Template... names the prefix and, when another prefix of the graph already stands for an address ending in that name - : for https://example.org/kg/ when the IDs say kg: - names it (since 1.134.0).
AL rdfs:label@en gives the row its name in English. (A rdfs:label@en looks the same and gives no name at all - Template... says so.)
Dates and numbers: write the type in the cell - 2026-11-01T18:00:00^^xsd:dateTime. ROBOT does not take it from the property, so a date typed plainly becomes text; Template... warns when the property says its values are dates.

Spreadsheets → Template... → Write me a template to start from writes a sheet with the columns most terms need, to copy.

2. Give every row an identifier¶

Leave the ID blank on new rows. Spreadsheets → Sheet editor opens the project's sheets by itself: the first of templates/ beside the open file, src/templates/, kg/templates/ and templates/ at the project root that holds a sheet - where it looks, exactly. The toolbar names the folder; Project templates ▾ offers the others. Select a sheet's tab and press Generate IRIs.... It offers the project's ID ranges, Continue the numbering in these sheets, a timestamp, a UUID or the label itself, previews what it will write, and never mints an identifier already used in a sheet of the folder, nor an IRI the ontology already uses. It compares identifiers as they are written: with CURIE IDs such as kg:PB_0000101 it cannot see the graph's own individuals, so do not delete a built row from its sheet and then generate again. Save writes the files.

3. Check the sheets¶

Spreadsheets → Check sheets... reads the whole folder against the open ontology and lists, cell by cell: names that point at nothing or at two things, a template placeholder left in a cell, pasted whitespace, a curly quote where a plain one was meant - each with the correction where there is an obvious one. Nothing is changed. The sheet editor marks the same problems where you type.

4. Build the graph¶

Open the graph file in Protégé. Spreadsheets → Template..., choose any sheet of the folder and Add every sheet in its folder: the sheets are read together, so their order does not matter, and Edit → Undo removes the lot in one step. Add the axioms to this ontology adds the one sheet - which can still name the rows of the others. A row that cannot be read in full is listed with its sheet, row and column. The other rows are built, and so is what that row says outside the failing cell - usually its ID and label - so run Check it, change nothing first, or use Edit → Undo after. Template... reads the saved sheet files, not what the sheet editor shows: save the sheets first. File → Save.

What Template... does, and what it does not. It turns the sheets into axioms and adds them to the ontology open in Protégé - in memory, until File → Save. It only ever adds: change a label in a sheet and run it again, and the graph has the old label and the new one. Nothing links the graph back to the sheets and nothing keeps them in step: an edit made to an individual in Protégé does not reach the sheet, and an edit to the sheet reaches the graph only when Template... is run again. To rebuild after changing rows, undo the earlier run (Edit → Undo, in the same session) or build into a fresh copy of the graph file - one that has its header, prefixes and imports and no individuals. Build and release → Build... builds the ontology, from the ontology's own window (from the graph's window it finds no Makefile), and the Makefile OntoBoard writes has no target for the graph: the graph is built with Template....

5. Ask it¶

Reasoner → HermiT → Start reasoner: whether the graph is consistent with the ontology, and what each thing turns out to be - a pizza with a hot topping is a spicy pizza when the ontology defines one, without anybody saying so.
Check and test → SPARQL...: questions - which pizzas each pizzeria baked on a day, who baked them, in which oven.

6. Version it¶

Give the graph a version IRI and a version - https://example.org/pizzeria/data/1.1.0 - and keep a copy of each release, as you do for the ontology. It imports the ontology release it was made for, by that release's version IRI, so a reader can tell exactly which ontology its data means.

Measured¶

On a test project: nine sheets - cities, pizzerias, bakers, their roles, ovens, instants, intervals, pizzas, bakings - with every name across sheets by label. The generator behind Generate IRIs... filled 16 blank identifiers, each sheet continuing its own numbering; Check sheets... had nothing to report; Template... added the nine sheets in order, 105 axioms and 25 individuals; HermiT found the graph consistent and one of the three pizzas spicy, from its type and the class hierarchy the ontology release states.