This paper describes the process of converting Cinema Context, an online dataset on Dutch lm culture, into Linked Open Data. It covers our experiences in this conversion process from the steps of data cleaning and modeling, up to publishing and evaluating the result through a case study. Cinema Context (CC) is an online encyclopedia on Dutch lm culture since 1896 [1]. Built on top of a MySQL database the website www.cinemacontext.nl o ers both an informational view as well as a research environment on places, persons and companies involved in more than 100k lm screenings in the Netherlands. The website allows a visitor to search and extract the data, though this is limited to the o ered capabilities of the faceted search. To provide access to the full dataset and to boost its interactivity, we have now published it as RDF. In this paper, we describe the process of converting a relational database with Cultural Heritage (CH) data into Linked Open Data (LOD). It also gives an example of the potential that this format o ers. Speci cally, converting this dataset to LOD brings opportunities for broadening and renewing historical and cultural research by allowing more exible linking to other (linked) datasets on for instance buildings, persons, heritage objects, and locations. Researchers in the Digital Humanities (DH) and CH communities have indicated a need [5] to be able to query CC in connection with external data via e.g. a SPARQL endpoint, in order to research the role of cultural and socio-economic status in processes of cultural consumption. Moreover, the selection of appropriate vocabularies and thesauri required close collaboration between data specialists and domain experts and has functioned as de facto training in working with RDF and the SPARQL query language for scholars working in DH.
The CC data model [
Place WikiData sameAs Country
location containedInPlace IMDB id
IMDB
Organization parentOrganization
parentOrganization worksFor MovieTheater organizer author
Person location
Event CC id
sameAs countryOfOrigin
publisher subEvent ScreeningEvent
Rating
sameAs contaebnotRuatting workPresented
Dutch National
Archives Movie isPartOf
Films and ratings We model each lm that is described in CC as a schema:Movie and provide information on its name, alternate name, production year, country of origin, format, and extent. A schema:sameAs property is available for every lm to refer to its entry on www.imdb.com. Ratings for these lms are modeled as schema:Rating and schema:CreativeWork and come from the archive of Film Screening Reports 1928-1960 which is held by the Dutch National Archives. The value in the schema:sameAs property of a rating points to the respective index in this collection.
Persons Persons in the CC data can be owners or employees of theaters and companies. They are modeled as schema:Person, provided with biographical information such as birth dates if available. A person's name is modeled through the Person Name Vocabulary (pnv)3.
Venues and companies Theaters or venues are organizations situated at a speci c location (schema:Place, including geometry) and with a speci c name. Usually, they are owned by a company or person. Each theater is a schema:Movie Theater, or a schema:EventVenue for general-purpose venues. If available, information on the seating capacity, number of screens, and lifespan is also given. The companies in this dataset are organizations that run cinemas or distribution companies and are modeled as an instance of schema:Organization. Events and screenings We distinguish two event classes: schema:Event for cinema programs, and sem:Event from the Simple Event Model (sem)4 for generic events, such as a venue's construction history. Instances of the rst consist of one or more schema:subEvents of type schema:ScreeningEvent ( lmic) or schema:TheaterEvent (non- lmic), held in a speci c theater on one or more dates, using a particular program name. A schema:startDate indicates the day a program started: normally, a cinema program would be screened for a week. Deviations from this norm are usually annotated as schema:description. 2.2
The schema.org vocabulary is complemented where it falls short, for instance when describing person names, legal entity types, and special cinema types (e.g. traveling cinema), but also when expressing a lm's length and extent, for which we use properties from Dublin Core5. Additionally, if a date is given in a less precise format than xsd:date, then the sem time stamp properties have been used to supply a proper date value. For consistency and usability, and to indicate a temporal restriction or (un)certainty, the sem properties are always present, even if an exact date is given.
This is also the case when resources are temporally restricted. We use the schema:Role class to express a speci c time frame in which a certain property value relation is valid. This class can be used in any object position and extends the triple with the same property, whilst incorporating additional information as quali er, such as speci c roles some entity played, or a start and end date. This way of modeling is used consistently in the data to boost its queryability, even when there is nothing to qualify. A description of less prominent auxiliary classes can be found in the dataset's documentation pages (see Section 3). 3
Documentation pages [
Both the documentation and the code that converts the MySQL database
are available in a git repository [
Case Study: International Orientation Index
A case study6 serves to illustrate the potential of connecting the CC dataset
with other knowledge graphs. It replicates the analysis of economic lm
historian Peter Miskell et al. [
We can test this hypothesis for the Dutch lm market by analyzing programming data from CC. What is missing in our dataset is information on box o ce revenues, but this value can be approximated by the number of screening weeks under the assumption that a lm with more screenings generates higher revenue. To approximate the variables Miskell et al. used to construct their index, we can apply the information available for lms in Wikidata. Instead of assigning a 0, 1 or 2 score to a criterion, we assigned a relative score (0.0-1.0) to the variables, indicating the extent of `internationalisation' (or rather: `non-Americanness') in a category. For a total of 8,836 lms (5,495 Hollywood productions), we gathered information on six categories through a SPARQL query, each retrieved with a particular Wikidata property path (e.g. the lm's director, followed by his/her country of citizenship). Examples of this calculation are shown in Table 1.
Calculating a correlation coe cient between the number of screenings and the relative internationalness of the Hollywood produced part of our corpus by using Pearson's r indicates that there is a very weak correlation of 0.130 when we consider lms (N=3418) for which we have information in at least three categories and 0.137 when we consider lms (N=1340) for which we have at least four categories (below three is not su ciently representing internationalness; over ve reduces the corpus size too much). Though positive, and thus indicating that internationally-oriented lms in the Netherlands perform slightly better than fully American ones, we should further re ne this proof of concept in future studies before making solid claims. 5
This project shows that the schema.org vocabulary can easily be applied to
cultural heritage data and deemed t for modeling our (research) dataset. With
6 A more detailed explanation of this and other case studies, including code and data,
can be found in the documentation pages [
The project was a collaboration between the CC editorial sta , Library UvA, and Menno den Engelse (Islands of Meaning).