Geographic place names are semantically often highly ambiguous. For example, there are 491 places in Finland sharing the same name "Isosaari" (great island) that are instances of several geographical classes, such as Island, Forest, Peninsula, Inhabited area, etc. Referencing unambiguously to a particular "Isosaari", either when annotating content or during information retrieval, can be quite problematic and requires usage of advanced search methods and maps for semantic disambiguation. Historical places introduce even more challenges, since historical metadata commonly make spatiotemporal references to historical regions and places using names whose meanings are non-existing or different in di erent times. This paper presents how these problems have been addressed in a large Finnish place ontology SUO and a historical geo-ontology SAPO. A location ontology server ONKI-Geo has been created for publishing the ontologies and utilizing them as mashup services. To demonstrate the usability of our ontologies, two case applications in the cultural heritage domain are presented.
Geospatial ontologies de ne classes and individuals for representing e.g.
geographic regions, their properties, and mutual relationships [
A problem when creating a semantic cultural heritage portal is that places, both modern and historical ones are needed but place nding and place name disambiguation is hard as there are millions of places. Moreover, geography changes rapidly, which makes it hard 1) for the content annotator to make correct references to spatiotemporal regions and 2) for the end-user to understand the changes in historical geography and, as a result, to formulate the queries. For example, many artifacts in MuseumFinland originate from regions that no longer exist and/or are not part of Finland but of Russia with new names after the Second World War. Finding the right names for querying, understanding to what regions on the map the names refer to at di erent times, and understanding how old historical names relate to modern Finnish and Russian geography creates, at the same time, both a semantic challenge for the technology and an important part of useful content to learn when using the portal.
We propose that (at least) three requirements should be taken into account when utilizing location information in a system for end-users to view cultural heritage collections: { Precise annotation with locations Finding relevant location or place concepts and their properties (like place type, WGS84 coordinates, hierarchical position) easily from geo-ontologies for annotating resources is a key need. { Ontology-based spatiotemporal search It is necessary to be able to use both historical and modern locations and regions as search concepts e.g. in a view-based, or multi-facet search. { Context visualization of the locations An end-user needs to see the locations in a right context, for example their neighbors, covered (i.e. overlapped) historical regions, and their position in a partonomical hierarchy or on a map.
To address these requirements, we present in this paper work done for 1)
modeling places, their properties and changes they have overcome as two ontologies
SUO and SAPO and 2) publishing the ontologies in an location ontology server
ONKI-Geo [
Spatial structures and their mutual relations are a most essential form of
geographic knowledge. The principles by which the geographic domain|and
geoontologies|are primarily structured from the theoretical viewpoint are topology
(the theory of boundaries, contact, and separation), mereology (theory of parts
and wholes) and geometry [
The SUO has been populated with 1) place information from the Geographic Names Register (PNR) provided by the National Land Survey of Finland7 and with 2) place information from the GEOnet Names Server (GNS)8 maintained by the National Geospatial-Intelligence Agency (NGA) and the U.S. Board on Geographic Names (US BGN). PNR contains about 800,000 names of natural and man-made features in Finland, including data such as place type or feature type and the coordinates of a place. The GNS register contains similar information of about 4,100,000 places around the world excluding places in the United States. An analysis of PNR and GNS produced the majority of classes of the SUO-ontology: together it contains over 800 di erent classes of man-made or natural geographical places, organized into a class hierarchy, and their topological and mereological relations, as well as relations de ning coordinate-based geometry for points, curves and polygons. SUO includes classes such as city, province, graveyard, lake, or river. 3
Names and borders for political geographic regions change over time: they are e.g. split (e.g. former Czechoslovakia) or merged together (e.g. former East and West Germany). Furthermore, also place names change (e.g. Zaire changed its name to Congo). As a result, content in databases may be indexed with names that do not exist anymore or refer to di erent regions than names in use today. This creates severe problems for information indexing and retrieval.
We have addressed the problem by developing a method and a historical place
ontology SAPO for representing geographical changes [
Figure 1 shows a part of SAPO where two regions, namely Nuijamaa (1944{1988) and Lappeenranta (1967{1988) have merged into one region called Lappeenranta (1989{). The merge is de ned as two RDF triples < A; mergedT o; C > and < B; mergedT o; C >, where A and B are regions that are merged to form C.
Based on the change types, logic rules can be applied to enrich the ontology with new useful relations. For example, in gure 1 the RDF statements covers 5 http://www.seco.tkk. /ontologies/suo/ 6 http://www.w3.org/2001/sw/ 7 http://www.maanmittauslaitos. /en/ 8 http://earth-info.nga.mil/gns/html/ and neighborOf can be derived based on the mergedTo change. A rule for deriving covers statements is given below: ! A covers C ^ C covers A ^ B covers C ^ C covers B
A mergedT o C ^ B mergedT o C
Another set of rules adds the triples < A; neighborOf; B > and < B; neighborOf; A > by checking if there are exactly two regions to be merged and if the area of the resulting region is made of a simple polygon.
Mutual covers relations of regions can be used to compute coverages between
regions of di erent times, without knowing the actual polygons of the regions,
and a geo-ontology time-series of regions [
The result can be used to show the temporal contextual information about places or in a query expansion. For example, to visualize the result of the inference and to see regions in relation with other regions, we have used the IRMASapo browser10, depicted in gure 2. With IRMA-Sapo browser, a user can nd historical regions beginning with certain letters and also their neighbouring regions, covering regions, and see them in their partonomy.
IRMA-Sapo is based on the intelligent metadata assistant IRMA [
The ONKI-Geo Browser is depicted in Figure 3. By using ONKI-Geo one can e.g. nd easily on the map all places of 1) some type with 2) names beginning with some letters, and that are 3) inside a polygon, out of millions of places12. ONKI-Geo uses Google Maps API for visualization, and can be connected to 11 http://ajax.org 12 A related service, Geonames.org also includes a web service to return place name information and supports search e.g. by type and name but not using e.g. a polygon. and utilized in legacy systems using Ajax and mash-up techniques. In a museum cataloging system, for example, places can be found by using the ONKI-Geo user interface and by pushing the \Select" button, the corresponding URI or coordinates are transferred from the centralized national service into the lecagy application. A technical challenge with the ONKI-Geo service is the large size of the SUO. The response times of the faceted search were several minutes when using a straight-forward Jena implementation 13. To solve the e ciency problem, only the SUO ontology classes, without the instances, are stored in a Jena model, and a separate indexing layer was created for the millions of instances. The indexing of the instances is created in a relational database (MySQL) making complex combinational searches possible and relatively fast.
We have implemented two case applications as a part of the semantic CultureSampo portal to demonstrate the use of place ontologies in the cultural heritage domain.
Map-search allows for searching with places on a map as illustrated in Figure 4. By clicking a place on a map, the items annotated with that place are retrieved and shown on the right side of the map. Map-search uses simple inference by creating a transitive closure over the part-of hierarchy.
Figure 5 illustrates the second case application. Historical regions of the
SAPO ontology can be selected in a drop-down menu on the left. Here a
temporal, historical place SAPO:Viipuri(1403-1943) is selected. As a result, the
polygonal boundaries of Viipuri of that time are visualized on a contemporary
Google Maps satellite image14, a map, or on a historical map. In addition, those
modern places of ONKI-Geo that are inside the polygonal boundaries of the
historical region are retrieved in a mashup fashion and can be used to browse the
map. The content related to SAPO:Viipuri(1403-1943) is retrieved from an RDF
repository and is shown in this view on the right. Furthermore, content from
historical regions that cover (i.e. overlap) SAPO:Viipuri(1403-1943) are listed as
recommendations. The coverages are looked up from the global coverage table
[
One of the major problems in semantic web ontologies is that organizations
commonly mint di erent URIs for same resources [
Our approach is to provide a rich set of spatial and temporal properties for the place instances in order to facilitate the process of disambiguating between them and for nding the needed instance (i.e. the URI) to be used e.g. in annotation. 14 http://maps.google.com/
We believe that if there is not enough evidence|i.e. properties and relationships between URIs|that a URI served by a service is the needed one both temporally and spatially then organizations tend to either mint their own URIs for the purpose or to select URIs somewhat randomly from di erent ad hoc -services.
This paper presented how two ontologies, SUO and SAPO, can be used to represent and visualize di erent kinds of locations and their changes in time. We also presented the ONKI-Geo ontology server for publishing SUO and SAPO, and for nding the correct places for annotations. Finally, we presented two case applications that demonstrate that the presented ideas work also in practise in the cultural heritage domain.