War history of the Second World War (WW2), humankind's largest disaster, is of great interest to both laymen and researchers. Most of us have ancestors and relatives who participated in the war, and in the worst case got killed. Researchers are eager to nd out what actually happened then, and even more importantly why, so that future wars could perhaps be prevented. The darkest data of war history are casualty records|from such data we could perhaps learn most about the war. This paper presents a model and system for representing death records as linked data, so that 1) citizens could nd out more easily what happened to their relatives during WW2 and 2) digital humanities (DH) researchers could (re)use the data easily for research.
Lots of information about the WW2 is available on the Web.1 However, this
information is typically meant for human consumption only. The underlying data
is not available in machine-readable, i.e., \semantic" form for Digital Humanities
research and use [
For the case of the First World War, the situation has started to change, with
several projects publishing linked data on the web, such as Europeana
Collections 1914{1918,2 1914{1918 Online,3 WW1 Discovery,4 Out of the Trenches,5
CENDARI,6 Muninn,7 and WW1LOD [
This paper discusses the publication and use of casualty (death) records as linked data, as one part of the larger WarSampo system. Here, a dataset of some 95,000 deaths in military action in the Finnish frontiers is concerned. We rst present the data, its modeling, and the Linked Open Data (LOD) service. After this two use case applications are presented: 1) Analyzing the data for DH research and 2) reassembling the biographical war history of individual soldiers and military units. The latter use case serves, e.g., laymen in trying to gure out what happened to their relatives in WW2. The WarSampo system9 was published on Nov 27, 2015 and has had tens of thousands of end users indicating a large public interest in such applications. 2
Information about all known Finnish casualties of WW2 has been gathered in a relational database at the National Archives. This database contains 94,696 records of persons that fought on the Finnish side, and died in 1939{1945 in the Winter War, the Continuation War, or in the Lapland War, or died of injuries obtained in those wars.
For use in the WarSampo project, the casualty database was rst converted to CSV format, which was then converted to RDF format. Because the objective was to develop interactive applications directly on top of the large RDF dataset, it was important to keep the amount of RDF triples as low as possible without losing information and still linking the death records to ontological concepts. Thus, a simple data model was created for representing the data as linked data.
The data model is based on representing each death record as an instance of
the Document class (E31 Document ) of the CIDOC Conceptual Reference Model
(CRM) vocabulary, which is designed for information exchange and integration
of various cultural heritage information [
A metadata schema was created that de nes the properties used to describe each casualty with the information from the original database. The schema consists of OWL properties which have crm:E31 Document as the domain. A list of the properties and their rdfs:range constraints are shown in Table 1. The namespace pre xes used in this paper are:
: http://ldf. /schema/narc-menehtyneet1939-45/ crm: http://www.cidoc-crm.org/cidoc-crm/ skos: http://www.w3.org/2004/02/skos/core# 8 http://www.bygle.net/wp-content/uploads/2015/04/Open-Memory-Project 31.pdf 9 Including a semantic portal in use at http://sotasampo. and the underlying LOD SPARQL service at http://www.ldf. /dataset/warsa/. wat: http://ldf. /warsa/actors/actor types/ wrank: http://ldf. /warsa/actors/ranks/
The default namespace corresponds to the casualty schema namespace. RDF Schema (RDFS), Web Ontology Language (OWL) and XML Schema namespaces are omitted. The properties are used only when there is a value for the property. Municipalities are currently linked to two distinct ontologies, which is why their range is not de ned.
Property description Property name Range mother tongue :aeidinkieli :Aeidinkieli occupation :ammatti xsd:string principal abode :asuinkunta
rst names :etunimet xsd:string date of becoming wounded :haavoittumisaika xsd:date municipality of becoming wounded :haavoittumiskunta place of becoming wounded :haavoittumispaikka xsd:string burial place :hautapaikka xsd:string burial graveyard :hautausmaa :Hautausmaa military unit :joukko osasto xsd:string military unit code :joukko osastokoodi xsd:string known military unit :osasto wat:MilitaryUnit citizenship :kansalaisuus :Kansalaisuus nationality at time of death :kansallisuus :Kansallisuus date of becoming missing :katoamisaika xsd:date municipality of becoming missing :katoamiskunta place of becoming missing :katoamispaikka xsd:string place of domicile :kotikunta date of death :kuolinaika xsd:date municipality of death :kuolinkunta place of death :kuolinpaikka xsd:string number of children :lasten lukumaeaerae xsd:integer perishing class :menehtymisluokka :Menehtymisluokka marital status :siviilisaeaety :Siviilisaeaety military rank :sotilasarvo wrank:Rank last name :sukunimi xsd:string gender :sukupuoli :Sukupuoli municipality of birth :synnyinkunta date of birth :syntymaeaika xsd:date full name skos:prefLabel rdfs:Literal WarSampo person instance crm:P70 documents crm:E21 Person
The Simple Knowledge Organization System (SKOS)10 was used to de ne
vocabularies to present the information found in the original database in RDF.
The created SKOS vocabularies for describing the death records in the casualty
dataset are listed in table 2.
The RDF dataset has been enriched by linking it to other parts of WarSampo
like military ranks, military units, information about persons found from other
sources, and municipalities of wartime Finland. The dataset is published in the
Linked Data Finland (LDF) [
A gure displaying the external linking of the death records is shown in Fig. 1. Each casualty is linked to four common WarSampo ontologies in addition to the vocabularies that are part of the casualty schema. Military ranks are linked to a common WarSampo military rank ontology according to the ranks of persons at the time of death. Persons that have been found and recognized from other sources are linked from the death records to their more general person instances via crm:P70 documents from their death records. New information about persons are not added directly to the casualty dataset, in order to maintain the integrity of the dataset as a whole.
The death records and military units are programmatically linked with
persons and military units found from other WarSampo sources. The linking is
implemented using the automatic annotation service ARPA [
For 118 death records, we were able to automatically link them to WarSampo persons from other sources. The amount is quite low because the person infor10 https://www.w3.org/2009/08/skos-reference/skos.html 11 http://www.ldf. /dataset/narc-menehtyneet1939-45 mation from other sources currently contains mostly information about high ranking o cers and persons who survived the war. The found person links have been manually validated and found links seem to be depicting the same persons. Manual validation was also done to person pairs that were close to the score threshold but not linked, and these seem to be either not depicting the same persons or don't have enough information to make an assumption either way.
Automatic linking to military units was implemented using the ARPA service. Some 66,700 death records we linked to WarSampo military units, so this counts for 70% of all the casualties.
Municipalities in the data are linked to the historical wartime municipalities
registry [
The dataset graph consists of almost 2.4 million RDF triples. Presenting the data in an online service for users to search and browse is not straightforward due to the large size of the dataset. Furthermore, there are lots of links to related data in other WarSampo datasets (persons, places, military units, etc.). 12 Described at http://www.ldf. /dataset/pnr
Faceted search provides e ective support for interactive information-seeking
in information systems [
Fig. 2 shows a screenshot of the faceted search application. The data is laid out in a table-like view. Facets are presented on the left of the interface with string search support. The number of hits on each facet is calculated dynamically and shown to the user, and selections leading to an empty result set are hidden. In addition, there is a text search facet for nding persons directly by name: the user just enters a person's name or a part of it into the search box, and also a date range selector to lter the results by date of death.
The faceted search is used not only for searching but also as a exible tool for researching the underlying data. In Fig. 2, the hit counts immediately show distributions of the killed widows along the facet categories. For example, the facet \Number of children" shows that one of the deceased had 10 children and most often (in 88 cases) widows had one child. If we next select the category \one child" on its facet, we can see that two of the deceased are women and 86 are men in the gender facet. 13 http://www.sotasampo. /casualties/
The application is developed in JavaScript as a Rich Internet Application
(RIA) on the client side, using the open SPARQL endpoint to fetch data
according to user selections. The application is open source,14 and is based on
our SPARQL Faceter tool [
The shown facets are con gured in the application and they directly use the properties of each death record instance. The property values are annotation resources selected from the corresponding ontologies, such as places.
The event perspective of the WarSampo portal also makes use of the casualty data as seen in Fig. 3. The application displays wartime events on a timeline and map. The casualty data is visualized by a heat map layer on the map, showing an overview of where casualties occurred during di erent time frames, and also which events happened nearby. The application also displays statistics regarding the casualties during the selected time frame: the total amount of casualties, and the amount per perishing class. Persons mentioned in the event descriptions are linked to the WarSampo person ontology, and thus to the casualties where the person has died in the wars. The application provides hyperlinks to the linked entities through their perspectives.
Fig. 4 shows a histogram of the amount of casualties per day of a single military unit, the 33rd infantry regiment, and all of its sub units. The time span 14 https://github.com/SemanticComputing/WarSampo-death-records covers the whole war, and shows also the war related deaths that occur after the war. Demonstrating the value of linking additional data to the death records, we have information of 7 events that are linked to the military unit and its sub units. They seem to explain quite well the casualties during the Winter War (30 November 1939 { 13 March 1940), as high peaks in casualty amounts occur when there are multiple concurrent battles. Information regarding this unit's activities during the Continuation War (25 June 1941 { 19 September 1944) are currently lacking. The unit and its sub units are known to have participated in the following events:
The histogram is created by reading data directly from the WarSampo SPARQL endpoint and visualizing it with YASGUI online SPARQL tool,15 and the following SPARQL query: PREFIX atypes: <http://ldf.fi/warsa/actors/actor_types/> PREFIX crm: <http://www.cidoc-crm.org/cidoc-crm/> PREFIX casualties: <http://ldf.fi/schema/narc-menehtyneet1939-45/> } ?cas casualties:osasto ?subunit .
?cas casualties:kuolinaika ?date . } GROUP BY ?date ORDER BY ?date SELECT ?date (count(?cas) as ?casualties) WHERE { { SELECT ?subunit
WHERE {
VALUES ?unit { <http://ldf.fi/warsa/actors/actor_1331> } . ?unit (^crm:P144_joined_with/crm:P143_joined)+ ?subunit .
?subunit a atypes:MilitaryUnit .
} } UNION {
VALUES ?subunit { <http://ldf.fi/warsa/actors/actor_1331> } . 4
Use Case 2: Reassembling Soldier Biographies and Military Unit Histories Linking the death records to information about the same persons in other sources, events, military units, war diaries, and wartime places provides new information about the persons involvement in the war events, his whereabouts and movements during the war. By linking all these pieces of information together, we are able to construct partial biographies of individual soldiers, and the movements 15 http://yasgui.org 120 90 60 30 Jan 1941
Jan 1942
Jan 1943
Jan 1944
Jan 1945 and actions of their military units. This allows an individual who is interested in investigating the biography of a relative who took part in the war to look at where the person probably fought, with whom, and when, and in what events his military unit participated. Also, the interlinked dataset, together with applications to e ectively use it, provides digital humanities researchers with new perspectives to study casualties, that would not be possible with a non-linked casualty dataset.
Fig. 5 depicts an integrated view of information related to a casualty in the persons perspective of the WarSampo portal. Basic information about a person (e.g., name, birth and death dates and places, occupation, marital status, military rank with promotion dates if available) is displayed on top of the page. After that, thumbnails of the linked photographs involving the person are shown. By clicking the thumbnails the user can explore the higher resolution versions of the photographs and their captions. In order to get further context to the person examined, the user can browse the army units the person belonged to during the war, and places related to his life events (e.g., birth and death municipalities on historical and contemporary maps). This way the user is able to track the person's participation in the war by investigating the movements of his army units and the durations of the battles the units fought.
The units perspective of WarSampo visualizes the troop actions both on a map and a timeline, and shows the casualties of the unit as a heat map in the same fashion as in the event perspective. For even deeper understanding of the history, links to digitized images of the war diaries of the army units are provided, containing rich primary source descriptions of the events. By following the municipalities links to the places perspective of WarSampo, one can, e.g., study what kind of war events took place in the person's birth place, see photographs taken at speci c locations the troops were located in, or read magazine articles depicting wartime events that took place in some place. 5
War history data is usually scattered in many isolated silos and may exist in totally di erent formats, e.g., books, paper archives, and databases. In this paper we have examined the bene ts and challenges of linking casualty records of war to multiple related datasets, and publishing them as linked data for DH research and applications to use. Two use cases were studied related to supporting DH research and services for the public.
In the future, linking with other WarSampo data will be developed further as new datasets are added to the whole. We plan to develop tools for statistical analysis of the data, and collaborate with humanities researchers in studying how linked data and our tooling can help to solve their research problems. An hierarchical occupation ontology is planned to be created and linked to the death records to provide insight into the social status of each casualty. Acknowledgements Tomi Ahoranta and Jeremie Dutruit from the National Archives contributed in publishing the original casualty database as LOD.
Our research is partially funded through the Finnish Open Science and Research Initiative by the Finnish Ministry of Education and Culture.