Second World War on the Semantic Web: The WarSampo Project and Semantic Portal

This paper initiates and fosters work on publishing Linked Open Data about the Second World War. It is argued that the heterogeneous, distributed data about the international world war history makes a promising use case for semantic technologies. We hope that by making war data openly available we can learn from the past and promote peace. 1 Publishing Linked Open Data about War History According to Georg Wilhelm Friedrich Hegel “we learn from history that we learn nothing from history”. Hopefully this is not the case for the Second World War (WW2), now that fighting has started again even within the borders of Europe in Ukraine. One way to promote peace is to make reliable data about the war openly available for everybody to learn. WarSampo is a project and semantic portal that aims at this goal by publishing large heterogeneous sets of data about the WW2 in Finland as Linked Open Data (LOD). Application demonstrators are built that provide different perspectives in war history, for both historians and the public. The data covers the Winter War 1939–1940 against the Soviet attack, the Continuation War 1941–1944 where the occupied areas of the Winter War were temporarily regained, and the Lapland War 1944–1945, where the Finns pushed the German troops away from Lapland. WarSampo1 is the next step in our series of “Sampo” portals based on Linked Data, including CultureSampo2 [9], BookSampo3, and TravelSampo4 and continues our earlier works on modeling the First World War [6,8]. The project started in autumn 2014 and is finished in 2017, by the centennial of Finland’s independence. 2 Data, Metadata Models, and Ontologies Data The project deals initially with the datasets presented in Table 1. The casualties data (1) includes data about the deaths in action during the wars. War diaries (2) are digitized authentic documentations of the troop actions in the frontiers. Photos and films 1 http://www.sotasampo.fi 2 http://www.kulttuurisampo.fi 3 http://www.kirjasampo.fi 4 http://www.seco.tkk.fi/projects/subi Dataset Name Providing organization Size 1 Casualties of WW2 National Archives 93,000 death records 2 War diaries National Archives 23,000 war diaries of troops 3 Photos & films Defence Forces & Military Museum 160,000 photos & films 4 Kansa taisteli magazine articles Bonnier & The Assoc. for Military History in Finland 3,360 articles of veteran soldiers 5 Karelian places National Land Survey 30,000 places of the annexed Karelia 6 Karelian maps National Land Survey War time maps of Karelia 7 Audio & films National Broadcasting Company YLE 250 recordings and films Table 1. Central datasets to be linked in WarSampo. (3) were taken during the war by the troops of the Defense Forces. The Kansa taisteli magazine (4) was published in 1957–1986; its articles contain mostly memories of the men that fought on the fronts. Karelian places (5) and maps (6) cover the war zone area in pre-war Finland that was finally annexed by the Soviet Union. YLE’s audio and film material (7) (“Living Archive”) was recorded during the war, or is related to it. Metadata Models CIDOC CRM5 is used as the harmonizing basis for modeling data, with events providing the semantic glue for data linking [3]. Our data model for WW1, presented in [8], is used as the metadata model to start with. Domain Ontologies The data is annotated using a set of domain ontologies, including: 1) an ontology of the troops and their hierarchies, 2) persons with their ranks and roles, 3) place ontology of historical places, 4) event ontology of battles, politics, and other war time incidents, 5) an ontology of time periods, 6) ontology of weapons, 7) ontology of vessels, and 8) a subject matter ontology. For 1–7 we have harvested named entities from the datasets, given them URIs and labels and some initial structure, as needed in our initial demos (discussed below). However, ontology modeling and development is still underway. A challenge of the actor ontologies, for example, is modeling the changes: names and positions of the troops as well as the roles of the personnel in the army change frequently (e.g., promotions of persons and changes in troop leadership) and have to be conditioned on time. For 8, the KOKO ontology, a center piece of the Finnish ontology infrastructure [4], is used. 3 Applications: Perspectives to War History The data and ontologies are published using SPARQL endpoints that form the basis of the WarSampo semantic portal and its applications. The idea of the portal is to provide a variety of different kind of perspectives to war data, represented on different tabs. Most 5 http://cidoc-crm.org Fig. 1. A heat map illustrating death counts on the map in WarSampo. datasets will have their own perspective, where the user can first search data of interest and then get linked data related to the resources found. The perspectives enrich each other via Linked Data. Initial prototypes for two perspectives have already been implemented: one for the war casualty data and one for the Kansa taisteli magazine articles. Fig. 1 depicts the user interface for the casualty data of 93,000 death incidents, with 6 facets on the left (marital status, gender, citizenship, nationality, mother tongue, and death category). On the top, an interactive timeline for the time facet is shown and below it there is a heat map illustrating the death counts on the maps during the selected time interval. Later on, death records will be enriched with links to, e.g., war diaries related to the dead person’s troop, related photos, and articles. The second demonstrator provides a faceted search interface to Kansa taisteli magazine articles, and links each article to further contextual data, such as related places, Wikipedia articles, troops, persons etc. based on the article metadata. Links to WarSampo demonstrators as well as further information about the project is provided at http://www.sotasampo.fi/en/. WarSampo is implemented using the “7-star” Linked Data Finland platform6 [7], based on Fuseki7 with a Varnish Cache8 front end for serving LOD. As a first official LOD publication, the casualty data from the National Archives is already publicly available for everyone to use9. 6 http://www.ldf.fi 7 http://jena.apache.org/documentation/serving data/ 8 https://www.varnish-cache.org 9 http://www.ldf.fi/dataset/narc-menehtyneet1939-45 4 Related Work and Discussion There are several projects publishing WW1 data on the web, such as Europeana Collections 1914–191810, 1914–1918 Online11, WW1 Discovery12, Out of the Trenches13, CENDARI14, Muninn15, and WW1LOD [8]. War history makes a promising use case for Linked Data because war data is heterogeneous, distributed in different countries and organizations, and written in different languages [5]. Many web sites publish data about the WW2. For example, the key datasets of WarSampo have been published in Finland by our collaborators, and in other countries many more sites are online, such as the World War II Database16 to name one. However, there are only few works on linking WW2 data, such as [2,1]. Much of the WW2 data is still confidential because people involved in the incidents or their close relatives are still alive. WarSampo contributes to related research by initiating and fostering large scale LOD publication of WW2 data, based on event-based data modeling. Our work is funded by the Ministry of Education and Culture and Finnish Cultural Foundation.