This paper describes the ontology model and published datasets of a digitized biographical person register. The applied ontology model is designed to represent people via their enduring roles and perduring lifetime events. The model is designed to support 1) prosopographical Digital Humanities research, 2) linking to resources in semantic Cultural Heritage portals, and 3) semantic data validation and enrichment by using SPARQL queries. The linked data approach enables to enrich a person's biography by interlinking it with space and time related biographical events, persons relating by social contacts or family relations, historical events, and personal achievements.
This resource description paper presents the LOD infrastructure, data model, and datasets used in the Norssit alumni register of short biographies. The data model is designed to support prosopographical research, data aggregation and linking in semantic portals, and easy SPARQL querying. The datasets consist of 10 137 person resources, enriched with graphs of relating career events and family relations, and vocabularies of titles, schools, companies, medals, and hobbies.
The data has been used in creating the Vanhat Norssit Portal3 allowing the user to
search and browse the data about individual persons as well as analyze and visualize
data about groups of people in proposographical research [
This paper is structured as follows: First, an ontology model for representing people with their life time events and relation roles is introduced. Secondly, the data sets of the use case Norssit alumni with information extraction from textual data is discussed. Then the results of entity linking are evaluated. Finally, the related work, lessons learned, and future work are discussed. 2
The ontology model representing people and their biographical information in the use
case Norssit alumni is based on the Bio CRM model7, which has been developed to
facilitate and harmonize the representation of biographies and cultural heritage data on
the Semantic Web. Bio CRM is a domain specific extension of CIDOC CRM8 [
Namespace http://ldf.fi/norssit/ http://ldf.fi/schema/bioc/ http://purl.org/dc/terms/ http://xmlns.com/foaf/0.1/ http://schema.org/ http://www.w3.org/2004/02/skos/core# Prefix : bioc: dct: foaf: schema: skos: rdf: rdfs:
The main classes of the person ontology are shown in Figure 1. A principle is that a foaf:Person instance only contains the properties that are considered constant, in our case family and first names, places, and dates of birth and death, etc. The person resource is enriched by attaching family relations, achievements, and titles. 7 http://seco.cs.aalto.fi/projects/biographies/ 8 http://cidoc-crm.org 9 http://ldf.fi/schema/bioc/
Family relation is an example of a binary, often even N-nary, relationship connecting two or more people. Social relationships can be modeled in a similar manner. Each family relation is a subclass of bioc:FamilyRelationshipRole. The domain specific ontology of family relations can build a hierarchy (e.g. :Mother and :Father are subclasses of :Parent). The RDF example below shows a definition of a class and declaration of a relationship between two people. Notice that the property bioc:inverse role has two values, depending on the gender of the relative. Genders and inverse relations can be used for data evaluation and reasoning: this SPARQL query10 fills in the missing inverse relationships in the dataset. A family relation is attached to a person using the property bioc:has family relation, which can have a blank node or a resource as a value. :Mother a rdfs:subClassOf bioc:inverse_role schema:gender skos:prefLabel :person_1 a schema:gender bioc:has_family_relation rdfs:Class ; :Parent ; :Son , :Daughter ; schema:Female ; "Mother"@en , "a¨iti"@fi . foaf:Person ; schema:Male ; [ a :Mother ; bioc:inheres_in :mother_1 ] .
A personal achievement refers to any notable activity (e.g. producing a work of art, a design project, receiving a political achievement, or participating in athletes games). Achievements are modeled so that a subproperty of :involved in connects the person to a instance of class :Achievement, and furthermore indicates what is the role of the person in a particular achievement. So a single achievement can refer to multiple people, and indicate the roles they participated with, e.g. an actor, author, or director of a movie. Instances of :Achievement can contain a description, information of time, place, and provenance, and link to a corresponding LOD resource or to a web page. :norssi_2230 :author :achievement_33 . :author a rdf:Property ; skos:prefLabel "teoksia"@fi , "Novels"@en ; rdfs:subPropertyOf :involved_in . :achievement_33 a skos:prefLabel dct:source :wikipedia <https://en.wikipedia.org/wiki/The_Egyptian> .
:Achievement ; "Sinuhe egyptila¨inen."@fi , "Sinuhe the Egyptian."@en ; <https://fi.wikipedia.org/wiki?curid=820> ; 10 Example of a SPARQL query: http://yasgui.org/short/BkRKKXyIZ
According to the Bio CRM principle, the occupation or profession of a person is considered a role. Another unary role modeled in a similar manner would be person’s nationality. The resource is a subclass of bioc:Title. The person involved is attached with a subproperty of bioc:inheres in, additional information like the company or medal of honour by subproperties of bioc:relates to. :tekniikan%20tohtori a rdfs:subClassOf skos:altLabel skos:prefLabel :event_21721 a bioc:inheres_in schema:startDate skos:prefLabel :tekniikan%20tohtori ; :norssi_7686 ; "1991"ˆˆxsd:gYear ; "TkT 91"@fi . 3
As a concrete case study, a register 1867–1992 of over 10 000 alumni of the prominent
Finnish high school “Norssi” was scanned, OCR’d, and transformed into RDF, then
enriched by data linking, published as a linked data service, and finally provided to end
users via a faceted search engine and browser for studying lives of historical persons
and for prosopographical research. [
The most important data source was the textual description of register entries. In Figure 2 a register entry is depicted, and some of the data fields are picked as examples. Description texts are well-formated, and always start with person’s name (a) and his place and date of birth (b, Jyva¨skyla¨, 19th Aug., 1868). Description includes names of person’s parents, and his years of enrollment and matriculation (c, yo 88). His later university degrees with graduation years are mentioned (d, FT = Ph.D.). The career is described as a list of entries in format Company role years (e, toimJ = CEO). Possible medals of honour are mentioned (f, VirVR 1 mk = Estonian Cross of Liberty, 1st Class) as well as military ranks with promotion years (g, Evl = Colonel Lieutenant). The description ends with a possible date of death (h, 11th Dec., 1939) and list of family relations (i, Veli = Brother). The data fields were extracted using regular expressions. 3.2
Currently the Norssit dataset contains ca 892 000 triples defining 131 000 resources. The main graphs are discussed in detail below, with the graph names, amounts of instances and triples, main classes, and properties.
Graph:
http://ldf.fi/norssit/people Contains: 17 791 instances, 183 972 triples Core classes: foaf:Person
This graph includes data of 10 137 people. A person resource contains biographical data extracted from register descriptions, e.g. given and family names, gender, places and dates of birth and death, years of enrollment, matriculation, or resignation, and provenance data. This graph also includes profile image URIs, family relationship instances, and links to external LOD clouds.
Graph: http://ldf.fi/norssit/events Contains: 34 000 instances, 247 237 triples Core classes: subclasses of bioc:Title
The event graph contains 34 000 career events extracted from register descriptions (e.g. see e in Fig. 2). Each resource contains links to an occupational or educational title, person involved, start and end years, a description, and a possible organization or medal. Altogether 5882 people are enriched with a title.
Graph: http://ldf.fi/norssit/achievements Contains: 3000 instances, 15 578 triples
The achievement graph contains 3000 personal achievements extracted from Wikipedia pages, or BookSampo11 Linked Data. In the case of Wikipedia, the information was extracted from HTML code under specified subtitles. Each resource provides a description, specified category, links to a Wikipedia page of the achievement, and link to person’s Wikipedia page served as an source of information.
Graph: http://ldf.fi/norssit/organizations Contains: 2300 instances, 5390 triples
The organization graph contains the labels and abbreviations of 2401 organizations, e.g. government agencies, companies, colleges, or universities. The labels were collected from text descriptions. An organization is attached to an event using the bioc:relates to property. Altogether 4805 people are linked to an organization.
Graph: Contains: 1760 instances, 3520 triples Core class: :Hobby
http://ldf.fi/norssit/hobbies The vocabulary of hobbies contains labels of 1760 different hobbies (e.g. Music, Sports, or Arts), mentioned in register descriptions. A hobby is attached using the :hobby property of Person resource. Altogether 7845 people are related with a hobby.
Graph: Contains: 350 instances, 1526 triples Core classes: subclasses of bioc:Title
http://ldf.fi/norssit/titles The titles graph contains classes of 350 educational or occupational titles and military ranks. These are the classes of instances in Career events graph. Altogether 5882 people have a specified title.
Graph: http://ldf.fi/norssit/medals Contains: 301 instances, 1254 triples Core class: :Medal
11 http://www.kirjasampo.fi The medals vocabulary contains 301 different types of honorary medals, the data is extracted from register descriptions (e.g. see f in Fig. 2). A medal is attached to an instance in Career events graph using the :relates to medal property. Altogether 1844 people are mentioned with a medal.
Graph: http://ldf.fi/schema/bioc Contains: 451 triples
This graph contains the definitions of the core classes and properties of the Bio CRM schema. It also includes the domain specific definitions of 45 subclasses of bioc:FamilyRelationshipRole (e.g. family members :Child, :Mother, or :Father) and 24 subproperties of :involved in (e.g. publication, design project, or nomination). 3.3
The Norssit data is linked to external data clouds shown in Table 2. The linkage was
done with string comparison using person’s full name with known dates of birth and
death. Links were created to Wikipedia, Wikidata, National Biography of Finland12
and its Swedish complement BLF13, BookSampo Linked Data, CultureSampo14 portal,
WarSampo15 portal, ULAN16 authority register by The J. Paul Getty Trust, VIAF17, and
the genealogical data service Geni18. For entity linking to databases offering a SPARQL
endpoint, the tool SPARQL ARPA19 was used. In cases where the database provides a
REST API, like Wikipedia or Geni.com, a special Python script was used. A Python
script was used also in the case of BLF, where the data was only available as a CSV
formatted table. [
The results of information extraction, and external data linking are evaluated in this chapter. Evaluation was done by first choosing an random sample at size of N = 50 or N = 100 people, and then manually checking if the data extracted or linkage accomplished was correct.
The results in Table 3 are all literal values of such properties that each person should have. Results indicate whether the information was interpreted correctly or not. For the 12 http://www.kansallisbiografia.fi/english 13 http://www.sls.fi/sv/projekt/blf-biografiskt-lexikon-finland 14 http://www.kulttuurisampo.fi 15 http://sotasampo.fi/en/ 16 http://www.getty.edu/research/tools/vocabularies/ulan/ 17 http://www.viaf.org 18 http://www.geni.com 19 http://seco.cs.aalto.fi/projects/dcert/
Geni 894 Wikipedia 609 Wikidata 609 CultureSampo 453 WarSampo 352 National Biography 136 VIAF 135 BookSampo 90 BLF 44 ULAN 16
Description Family research and family tree data http://fi.wikipedia.org http://www.wikidata.org LOD from museums, archives, libraries, and media Second World War LOD service and portal National Biography of Finland Virtual International Authority Files Finnish fiction literature on the Semantic Web service Biografiskt Lexikon fo¨r Finland
Union List of Artist Names Online property of name, the single false result was caused by an error in the OCR process, which caused an erroneous family name in the dataset. In the register book, the dates of birth and death were written in format dd MM yyyy with the month in Roman numerals (see b and h in Fig. 2). The two false results were caused by a typical OCR problem of mixing up characters l, 1, and I. The year of enrollment was annotated in two digit decade and year format (c in Fig. 2, e.g. 83 for 1883 or 1983), and the century was automatically reasoned based on the person’s birth year.
The Norssi high school has had female pupils only after the year 1972; approximately 11 per cent of people in the data set are female. For pupils enrolled 1972 or after, the gender was generated in three steps. First, depending on the known family relations, some people were marked male or female. Next, given names of people remaining without a specified gender were compared to the given names of people with known gender, and people with matching names inherited the corresponding gender. Finally, a list with less than 100 otherwise unidentified rare names was filled manually.
Description Name Date of birth Year of enrollment Gender
Property skos:prefLabel schema:birthDate :enrollmentYear schema:gender
Correct
False
Precision
Unlike in the previous examples, the properties evaluated in Table 4 were not obligatory. So, like for instance for the date of death there are 10 true positive (TP) matches where the information was interpreted correctly, 38 true negative (TN) cases where the data had no such a date, one false positive (FP) case caused by noise in the OCRprocess, and one false negative (FN) where the information was not retrieved. The year of matriculation was an integer extracted from the text just like the year of enrollment. Profile images for each corresponding person (see e.g. Fig. 2) were located from the OCR’d layout in the XML-format.
Some of the properties were very sparse in data, so the sample size was increased to N = 100 for the evaluation of hobbies, family relations, careers, and links to LOD cloud. The false results in the family relations were caused by misinterpreting certain words of the Finnish language (e.g. the word Eno (Uncle) is also a name of a village). The data of the family relationships was further evaluated with SPARQL queries checking some basic rules, e.g. a parent must be older than the child20.
The algorithm for linking to external databases (Wikipedia, Semantic National
Biography of Finland [
FP TN FN
Our research group, Semantic Computing Research Group (SeCo) has produced several
projects with actor ontologies: The National Biography of Finland, CultureSampo21,
20 Example of a SPARQL query: http://yasgui.org/short/rJI8CXyUb
21 http://seco.cs.aalto.fi/applications/kulttuurisampo/
BookSampo22, and WarSampo [
CIDOC CRM includes a mechanism for representing the role of an active
participant in an event, modeling it as a property of the property that states the participant (see
CIDOC’s P14.1 in the role of). There is a proposal for encoding CIDOC’s properties
of properties as RDF23, introducing new class for the property and auxiliary properties,
which adds complexity to the data model. Simple Event Model (SEM) [
Bio CRM’s approach aims for simplicity and compatibility with CIDOC CRM. The model supports expressing enduring unary roles and binary relationships without the need to model them in the context of an event. 5.2
In our dataset, some practices were simplified, like modeling of a person’s birth with literal values of properties :birth place and :birth time instead of using the CIDOC CRM event crm:E69 Birth. These simplifications worked well in this case, and reduced the complexity of the data. Another similar case was modeling people’s names as literal values instead of using a resource of the type crm:E41 Appellation. 5.3
We will continue our work on modeling and publishing biographies with data publications dealing with Finnish Biographies25, and U.S. Congress Legislators26.
HISCO27 [