Theatre Migrants

To generate a knowledge graph about migrants in the theatre in Europe.

Running the scripts

The mapping scripts have been reimplemented in Rust for faster execution. Both scripts must be run from this directory (mapping/).

Prerequisites: Start the MariaDB container before running step 1:

docker compose up -d

Step 1 — Direct Mapping from MariaDB to RDF (data/graph-01.ttl):

cargo run --release --bin step-01

Step 2 — Apply SPARQL UPDATE queries (data/graph-02.ttl):

cargo run --release --bin step-02

Alternatively, after installing with cargo install --path .:

step-01
step-02

Generating the ontology

Next there are set of steps describing how to generate the migrants RDF graph.

Step 1 - Loading the input data into a relational database

Task

The file teatre-migrants.sql contains the dump of a MariaDB database. The tables involved in this schema are described in the file db_schema.md. We will load this data in MariaDB to access the data with SQL. To this end:

1. Create a Dockerfile to create a docker container for MariaDB.

2. Upload the dump into a database in the container.

3. Create a Rust program src/map/step_01.rs that connects to the database. This program should return a file called graph-01.ttl containing all the data from the tables loaded in the database using the direct mapping from relational databases to RDF.

Summary

The Dockerfile creates a MariaDB 10.11 container that automatically loads teatre-migrants.sql on first start. The docker-compose.yml exposes the database on port 3306 with a healthcheck.

The program src/map/step_01.rs connects to the database and implements the W3C Direct Mapping for all 9 tables (location, migration_table, organisation, person, person_profession, personnames, relationship, religions, work). Each table row becomes an RDF resource identified by its primary key, each column becomes a datatype property, and each foreign key becomes an object property linking to the referenced row. The output file graph-01.ttl contains 162,029 triples.

To run:

docker compose up -d
cargo run --release --bin step-01

Step 2 - Generate Objects

Continents and countries should be objects instead of literals. To this end, we can transform the following data:

base:location\/ARG-BahBlanca-00 a base:location;
  base:location\#City "Bahia Blanca";
  base:location\#Continent "South America";
  base:location\#Country "Argentina";
  base:location\#GeoNamesID "3865086";
  base:location\#IDLocation "ARG-BahBlanca-00";
  base:location\#latitude -3.87253e1;
  base:location\#longitude -6.22742e1;
  base:location\#wikidata "Q54108";
  base:location\#wikipedia "https://en.wikipedia.org/wiki/Bah%C3%ADa_Blanca" .

Into the following data:

base:location\/ARG-BahBlanca-00 a base:location;
  base:location\#City base:City-BahiaBlanca;
  base:location\#Continent base:Continent-SouthAmerica;
  base:location\#Country base:Country-Argentina;
  base:location\#GeoNamesID "3865086";
  base:location\#IDLocation "ARG-BahBlanca-00";
  base:location\#latitude -3.87253e1;
  base:location\#longitude -6.22742e1;
  base:location\#wikidata "Q54108";
  base:location\#wikipedia "https://en.wikipedia.org/wiki/Bah%C3%ADa_Blanca" .

base:City-BahiaBlanca a base:City;
  rdfs:label "Bahia Blanca"@en .

base:Continent-SouthAmerica a base:Continent;
  rdfs:label "South America"@en .

base:Country-Argentina a base:Country;
  rdfs:label "Argentina"@en .

Notice that all ranges of property rdfs:label are stated to be in English.

Generate an SPARQL UPDATE query that do this tranformation for all elements of the table and save it a new folder called updates. Do the same with the other tables, proposing which columns should be defined as objects. For every table define a different SPARQL UPDATE query and to be saved in the updates folder. Enumerate these generated queries adding a prefix number like 001, 002, 003, and so on.

After generating the update queries, generate a Rust program that executes the updates on the RDF graph generated in the previous step and generates a new RDF graph to be saved: data/graph-02.ttl.

Summary

19 SPARQL UPDATE queries in updates/ transform literal values into typed objects across all tables:

Query	Table	Column	Object type
001	location	Continent	Continent
002	location	Country	Country
003	location	State	State
004	location	City	City
005	migration_table	reason	MigrationReason
006	migration_table	reason2	MigrationReason
007	organisation	InstType	InstitutionType
008	person	gender	Gender
009	person	Nametype	Nametype
010	person	Importsource	ImportSource
011	person_profession	Eprofession	Profession
012	personnames	Nametype	Nametype
013	relationship	Relationshiptype	RelationshipType
014	relationship	relationshiptype_precise	RelationshipTypePrecise
015	religions	religion	Religion
016	work	Profession	Profession
017	work	Profession2	Profession
018	work	Profession3	Profession
019	work	EmploymentType	EmploymentType

Each query replaces a literal value with an object reference and creates the object with rdf:type and rdfs:label (in English). The program src/map/step_02.rs loads data/graph-01.ttl, applies all queries in order, and writes data/graph-02.ttl (164,632 triples).

To run:

cargo run --release --bin step-02

Step 3 - Annotate dataypes

In the previous example we have dates like "1894-12-31", which is represented as an xsd:string datatype. Please infer the datatypes of these literals and create a new SPARQL query to generate a new RDF graph where literals use these dataypes.

Step 4 - Replace empty string with unbound values

Intuitively, the triple

work:4 workp:EmploymentType workp:comment "" .

does not intended to mean a comment "", but the lack of a comment. So, write a query that exclude these comments from the next generated graph.

Step 5 - Use well-known vocabularies

For some classes, properties, and individuals we can be represented with Schema.org. For example, the class migrants:person can be represented with the class schema:Person. Please propose what of these elements could use the Schema.org vocabulary and generate an SPARQL to generate the next graph. Consider using other vocabularies beyond Schema.org, if you consider them appropiate to represent the information on this dataset.

Summary

7 SPARQL UPDATE queries in updates_step05/ add well-known vocabulary properties alongside the existing migrants: predicates:

Query	Mapping
001	Person properties → schema:givenName, schema:familyName, schema:birthDate, schema:deathDate, schema:gender, schema:birthPlace, schema:deathPlace, schema:image, schema:hasOccupation, schema:citation, rdfs:comment
002	Person authority identifiers (Wikidata, GND, VIAF, CERL, LCCN, ISNI, SNAC) → owl:sameAs and wdtn: normalized properties
003	Location properties → wgs84:lat, wgs84:long; Wikipedia/Wikidata links → owl:sameAs
004	Organisation properties → schema:name, schema:location, rdfs:comment
005	Person labels → rdfs:label (generated from first_name + family_name)
006	Enumeration instances → skos:Concept + skos:prefLabel
007	Class types → schema:Person, schema:Place, schema:Organization

The program src/map/step_05.rs loads data/graph-04.ttl, applies all queries, and writes data/graph-05.ttl (168,129 triples).

To run:

cargo run --release --bin step-05

Step 6 - Map to the Theatre Migrants ontology

Task

Define a custom OWL ontology (teatre-migrants.ttl) for domain-specific terms not covered by well-known vocabularies, published at https://daniel.degu.cl/ontologies/theatre-migrants/ with prefix tm:. Reuse existing vocabularies where possible:

• Schema.org for persons, places, organizations, and occupations.
• W3C Organization Ontology (org:) for work engagements, modeled as org:Membership (replacing the original migrants:work class). Properties org:member and org:organization link the membership to the person and organization.
• SKOS for enumeration types as subclasses of skos:Concept.

Write SPARQL CONSTRUCT queries that produce a new graph using only the tm:, schema:, org:, skos:, owl:, wgs84:, and wdtn: vocabularies. The original http://example.org/migrants/ predicates and class types are replaced; only entity IRIs retain the migrants: namespace.

Summary

The ontology teatre-migrants.ttl defines:

• 5 domain-specific classes: tm:Migration, tm:Relationship, tm:PersonName, tm:ReligionAffiliation, tm:ImportSource (tm:PersonProfession was removed in Step 7).
• 11 enumeration classes (all rdfs:subClassOf skos:Concept): tm:Continent, tm:Country, tm:State, tm:City, tm:MigrationReason, tm:InstitutionType, tm:NameType, tm:RelationshipType, tm:RelationshipTypePrecise, tm:Religion, tm:EmploymentType.
• Object and datatype properties with domains, ranges, and temporal uncertainty modeling (tm:dateStartMin, tm:dateStartMax, tm:dateEndMin, tm:dateEndMax, tm:dateStartFuzzy, tm:dateEndFuzzy).

12 SPARQL CONSTRUCT queries in constructs_step06/ transform the graph:

Query	Description
001-persons	Persons with schema:Person properties and tm: extensions
002-places	Places with wgs84: coordinates and tm: geographic hierarchy
003-organisations	Organizations with schema:name and tm:institutionType
004-migrations	Migration events with tm:migrant, tm:startPlace, tm:destinationPlace
005-memberships	Work engagements as org:Membership with org:member, org:organization
006-relationships	Interpersonal relationships with tm:activePerson, tm:passivePerson
007-person-professions	Person–profession associations
008-person-names	Historical/alternative person names
009-religion-affiliations	Religion affiliations with temporal bounds
010a-occupations-passthrough	Pass through existing schema:Occupation instances
010b-occupations-from-profession	Retype migrants:Profession as schema:Occupation
011-enumerations	Map enumeration instances to skos:Concept with tm: subtypes

The program src/map/step_06.rs loads data/graph-05.ttl, runs all CONSTRUCT queries, collects the resulting triples into a new graph, and writes data/graph-06.ttl (148,985 triples).

To run:

cargo run --release --bin step-06

Step 7 - Clean up secondary organisations and simplify person–profession

Task

Two clean-up tasks are performed on the graph produced by Step 6:

Secondary organisations. org:Membership instances may carry a tm:secondaryOrganisation property in addition to org:organization. An analysis of the 1,222 memberships with a secondary organisation reveals:

Category	Count
Secondary differs from primary	736
Secondary equals primary (redundant)	230
Secondary exists but no primary	256
Total with secondary organisation	1,222

Two SPARQL UPDATE queries clean up these cases:

1. Remove redundant secondary — when tm:secondaryOrganisation equals org:organization, delete the secondary (230 triples removed).
2. Promote secondary to primary — when a membership has tm:secondaryOrganisation but no org:organization, move the secondary to primary (256 triples replaced).

After these updates, 736 memberships retain a tm:secondaryOrganisation that genuinely differs from the primary organisation.

Person–profession simplification. The tm:PersonProfession class modeled an intermediate node linking persons to professions (from the person_profession database table). Since both the profession and Eprofession columns represent occupation names (schema:name), the intermediate class is replaced by direct schema:hasOccupation links from persons to schema:Occupation instances. The tm:PersonProfession class and its properties (tm:personProfessionPerson, tm:enumeratedProfession, tm:professionLabel) are removed from the ontology.

Summary

5 SPARQL UPDATE queries in updates_step07/:

Query	Description	Affected
001	Remove tm:secondaryOrganisation when it equals org:organization	230
002	Promote tm:secondaryOrganisation to org:organization when no primary exists	256
003	Add schema:hasOccupation from person to enumerated profession	3
004	Create schema:Occupation from profession label and add schema:hasOccupation	730
005	Remove all tm:PersonProfession instances	742

The program src/map/step_07.rs loads data/graph-06.ttl, applies all queries, and writes data/graph-07.ttl (147,431 triples).

To run:

cargo run --release --bin step-07

Step 8 - Merge migration reasons and refactor temporal properties

Task

Two structural changes are applied to the graph produced by Step 7:

Merge migration reasons. The functional properties tm:reason (774 uses) and tm:secondaryReason (163 uses) are replaced by a single non-functional property tm:hasReason, resulting in 937 reason triples.

Refactor temporal properties into tm:FuzzyInterval. Six flat date properties (tm:dateStartMin, tm:dateStartMax, tm:dateEndMin, tm:dateEndMax, tm:dateStartFuzzy, tm:dateEndFuzzy) are replaced by a structured model based on W3C OWL-Time. A new class tm:FuzzyInterval (subclass of time:TemporalEntity) is introduced, with two object properties tm:uncertainBeginning and tm:uncertainEnd pointing to time:DateTimeInterval resources. Each interval has time:hasBeginning and time:hasEnd linking to time:Instant nodes with time:inXSDDate values, plus an optional rdfs:label for fuzzy date strings. Five classes are declared as subclasses of tm:FuzzyInterval: tm:Migration, org:Membership, tm:Relationship, tm:PersonName, tm:ReligionAffiliation.

Summary

8 SPARQL UPDATE queries in updates_step08/:

Query	Description
001	Merge tm:reason and tm:secondaryReason into tm:hasReason
002a	Create tm:uncertainBeginning interval from tm:dateStartMin
002b	Add upper bound to tm:uncertainBeginning from tm:dateStartMax
003a	Create tm:uncertainEnd interval from tm:dateEndMin
003b	Add upper bound to tm:uncertainEnd from tm:dateEndMax
004a	Add rdfs:label on uncertainBeginning from tm:dateStartFuzzy
004b	Add rdfs:label on uncertainEnd from tm:dateEndFuzzy
005	Remove all 6 old date properties

The program src/map/step_08.rs loads data/graph-07.ttl, applies all queries, and writes data/graph-08.ttl.

To run:

cargo run --release --bin step-08