This paper shows how various prosopographical phenomena can be highlighted and visualized in the WarSampo Knowledge Graph that contains rich data about Finland in the Second World War as Linked Open Data, including detailed metadata of more than 100 000 people. WarSampo Portal contains tools for simple prosopographical data analysis of the person registers, and accessing the SPARQL endpoint directly opens up further possibilities in using the ontology infrastructure for enhanced information retrieval and pursuing digital humanities studies. This paper overviews of these possibilities of WarSampo, and presents examples of how it, and by extension Linked Data more generally, can be used to created data analyses to support historical research.
for external tools. Just a SPARQL query and result set visualizations already enable answering complex questions about history. For example, one can study how the ratio of oficers in perished soldiers difered geographically. The portal provides nine interactive “perspectives” on the data: (War) Events, Persons, Army Units, Places, Magazine Articles, Casualties, Photographs, War Cemeteries, and Prisoners of War. Since its opening in 2015, the WarSampo portal has been used by more than a million end users, corresponding to almost 20% of the population of Finland.
Semantic Web technologies provide viable solutions for combining heterogeneous isolated
historical datasets [
The WarSampo KG enables seeking new insights about WW2, the arguably most devastating
catastrophe in human history. The ontology and data infrastructure of WarSampo can be further
extended with new data to enable digging even deeper into the societal research questions
which interest many military history scholars today [
This paper extends our earlier publications on WarSampo by showing how the LOD service and the portal can be used in Digital Humanities data analyses. This paper shows how various prosopographical phenomena can be highlighted and visualized. For example, we show 1) how the data can be used to visualize geographical variance in the ratio between enlisted soldiers and oficers, 2) how the perished soldiers’ places of domicile correlate with their mortality, social class, and place of burial, and 3) how one’s occupation afected the likelihood of surviving in POW camps.
In WarSampo, Linked Data and the event-based CIDOC Conceptual Reference Model (CRM) [
The death record data is of great importance in studying Finland in WW2, as it contains detailed information about all 94 700 perished soldiers in the Finnish fronts. The POW register contains data about all 4200 Finnish POWs. In addition, WarSampo contains information of over 5600 notable persons who survided the war, aggretaged from additional data sources.
The occupations of the person registers have been harmonized into an occupation
ontology [
There are also person related documents that are linked to the person instances or their
military units, including a large collection of wartime photographs, hand-written digitized
war diaries, and war veteran magazine articles. These provide further contextual information
for people studying, for example, the war paths of their relatives. The latest version of the
WarSampo KG is always available at the LDF.fi platform 3. All versions are available from
Zenodo, and the current 2.1.0 version [
Using Semantic Web technologies and CIDOC CRM help to create a sustainable and collaborative
infrastructure for pursuing historical research [
Many of the domain ontologies of WarSampo, e.g., military ranks and war-time municipalities, are used to provide facet values in the many faceted search perspectives of WarSampo Portal. These can be re-used to enable faceted search with other datasets.
The WarSampo infrastructure has recently been employed in the WarMemoirSampo system4 to provide contextual information to the things being discussed in war veteran interview videos, such as places, organizations, persons, military units, and events.
A number of data analytical visualizations can be performed on the WarSampo portal [
Figure 1 presents a screenshot of the soldier life paths as a Sankey diagram from the Casualties
perspective [
Cemeteries, local histories, and family histories interest the wide public to study the provided historical information to find information related to their own lives. In addition academic researchers and military history enthusiasts study the data through the portal.
To go beyond the possibilities of the semantic portal, one can use the LOD directly from the underlying SPARQL endpoint hosting the KG. LOD makes it potentially much easier to share research data in a way that benefits all. The main drawback is the new technical skills that need to be acquired. This section shows some examples of how visualizations can be created for prosopographical analyses using the WarSampo KG.
SPARQL query language is a major tool when using LOD. Below is an example of a relatively
simple query5 that will count the number of deaths among farmers, fishermen and others, based
on the HISCLASS classification of their occupations. At the start of the query there is a number
of PREFIX definitions that are not strictly necessary, but make the query easier to read and
write. For example, the URI of the death record class http://ldf.fi/schema/warsa/DeathRecord
can be written as warsa:DeathRecord after defining the prefix warsa. The SELECT line indicates
the variables (identified by ’ ?’ character before variable name) that we want to extract from
the data, in this case date of death and the number of deaths (per day). The section after
WHERE include triple patterns that limit the query to the group that we are interested in, in
this case people with HISCLASS 5-class scheme class 3 (farmers) who died during the Winter
War. For example, the line ’?record a warsa:DeathRecord .’ is a triple pattern; it
means that we want the variable ’?record’ to represent all the resources in the KG that are of
type http://ldf.fi/schema/warsa/DeathRecord . We then limit the set of death records based on
occupation and death date. At the end of the query we sort the results based on the dates in
ascending order. These results that consist of dates and numbers can be visualized, e.g., as a
line chart shown in Figure 2 created with the YASGUI editor tool [
5Yasgui can be used to share SPARQL queries; the query is here: https://api.triplydb.com/s/DCbeABJmE
PREFIX xsd: <http://www.w3.org/2001/XMLSchema#> PREFIX warsa: <http://ldf.fi/schema/warsa/> PREFIX casualties: <http://ldf.fi/schema/warsa/casualties/> PREFIX bioc: <http://ldf.fi/schema/bioc/> PREFIX ammo: <http://ldf.fi/schema/ammo/> SELECT ?date_of_death (COUNT(?record) AS ?number_of_deaths) WHERE { ?record a warsa:DeathRecord . ?record bioc:has_occupation ?occupation . ?occupation ammo:hisclass5 <http://ldf.fi/ammo/hisco/hisclass5/3> . ?record warsa:date_of_death ?date_of_death .
FILTER (?date_of_death >= "1939-11-30"^^xsd:date)
FILTER (?date_of_death <= "1940-03-11"^^xsd:date) } GROUP BY ?date_of_death ORDER BY ASC(?date_of_death)
Often even somewhat complex visualizations can be created quickly with essentially only a SPARQL query. An example of this is can be seen in Figure 3. The bar chart visualization shows the proportion of perished oficers to all perished soldiers in the data for each Finnish province based on the municipality of residence of each victim. It is easy to see that the Uusimaa Province (“Uudenmaan lääni”) , where the Finnish capital Helsinki is located, has considerably higher ratio of oficers than the other provinces.
The visualization was created using the visualization tools of Yasgui and a short SPARQL query6 of only a few lines. The places and ranks have simple hierarchical ontologies that are used to determine the province of residence and if the victim is an oficer or not. This visualization uses only Finnish labels for the provinces as retrieved from the KG. However, LOD makes it easy to have names of entities in multiple languages using language tags.
The register of the Finnish POWs is much smaller than the casualty register, but comparative studies with it can still provide interesting insights. Figure 4 shows the 7-class HISCLASS distribution of the prisoners in two groups: 1) those who survived the POW camps (blue), and 2) those who didn’t survive (red), visualized with the Yasgui tool7. The large group “No HISCLASS code” corresponds mostly to non-specialized workers who worked in various tasks, often seasonally (“työmies” and “sekatyömies” in Finnish). There are 2726 prisoners with known occupation who survived and correspondingly there are 1393 perished prisoners.
One can see that there are diferences in the distribution between the two groups in Figure 4. One can study this further by inspecting the group with the largest diference (omitting the “No HISCLASS code”): “Foremen and medium skilled workers”. A visualization8 of the 10 most common occupations in this group is shown in 5. This suggests that skilled workers had better chances of survival than others. As suggested by a collaborating historian, some skilled workers
have been valued during their imprisonment as they have been useful in practical tasks, such as carpentry, and therefore might have received better treatment.
In this paper we have shown how LOD can facilitate research in military history, as exemplified with the WarSampo KG. Metadata that is harmonized and reconciled into ontologies, like in the ontology infrastructure of WarSampo, provide enhanced possibilities for information retrieval that, combined with data analysis, can be fruitful for comparative prosopographical studies. SPARQL is a potent tool for such studies once a suitable Knowledge Graph is available, but it is a tool that takes some efort to master for a person without computational skills.
To go beyond the portrayed examples, it would be crucial to pursue interdisciplinary Digital Humanities collaboration between Semantic Web researchers and humanist scholars. The former would provide the technical skill set needed for the studies, and the latter would provide the historical understanding of what would be interesting to study and how to interpret the produced results by setting them in context.
LOD enables new kinds of Digital Humanities studies by making it possible to easily do analysis that would be laborious, and often not feasible, to do otherwise. To use full potential of the possibilities required developing an ontology infrastructure to which the data is linked. LOD is most useful when integrating a number of heterogeneous, naturally interlinked datasets, providing a solid foundation for building such data infrastructures, that can be accessed and used by others.
We wish to acknowledge CSC – IT Center for Science, Finland, for computational resources.