Rome in three days, an archivists introduction to linked data publishing
If you to go to Rome for a few days, do everything you would do in a single day, eat and drink in a few cafes, see a few fountains, and go to a museum of your choice.
Linked data in archival practice is not new. Others have been here previously. You can benefit from their experience and begin publishing linked data right now using tools with which you are probably already familiar. For example, you probably have EAD files, sets of MARC records, or metadata saved in database applications. Using existing tools, you can transform this content into RDF and put the result on the Web, thus publishing your information as linked data.
If you have used EAD to describe your collections, then you can easily make your descriptions available as valid linked data, but the result will be less than optimal. This is true not for a lack of technology but rather from the inherent purpose and structure of EAD files.
A few years ago an organisation in the United Kingdom called the Archive’s Hub was funded by a granting agency called JISC to explore the publishing of archival descriptions as linked data. The project was called LOCAH. One of the outcomes of this effort was the creation of an XSL stylesheet (ead2rdf) transforming EAD into RDF/XML. The terms used in the stylesheet originate from quite a number of standardized, widely accepted ontologies, and with only the tiniest bit configuration / customization the stylesheet can transform a generic EAD file into valid RDF/XML for use by anybody. The resulting XML files can then be made available on a Web server or incorporated into a triple store. This goes a long way to publishing archival descriptions as linked data. The only additional things needed are a transformation of EAD into HTML and the configuration of a Web server to do content negotiation between the XML and HTML.
For the smaller archive with only a few hundred EAD files whose content does not change very quickly, this is a simple, feasible, and practical solution to publishing archival descriptions as linked data. With the exception of doing some content negotiation, this solution does not require any computer technology that is not already being used in archives, and it only requires a few small tweaks to a given workflow:
- implement a content negotiation solution
- create and maintain EAD file s
- transform EAD into RDF/XML
- transform EAD into HTML
- save the resulting XML and HTML files on a Web server
- go to step #2
EAD is a combination of narrative description and a hierarchal inventory list, and this data structure does not lend itself very well to the triples of linked data. For example, EAD headers are full of controlled vocabularies terms but there is no way to link these terms with specific inventory items. This is because the vocabulary terms are expected to describe the collection as a whole, not individual things. This problem could be overcome if each individual component of the EAD were associated with controlled vocabulary terms, but this would significantly increase the amount of work needed to create the EAD files in the first place.
The common practice of using literals to denote the names of people, places, and things in EAD files would also need to be changed in order to fully realize the vision of linked data. Specifically, it would be necessary for archivists to supplement their EAD files with commonly used URIs denoting subject headings and named authorities. These URIs could be inserted into id attributes throughout an EAD file, and the resulting RDF would be more linkable, but the labor to do so would increase, especially since many of the named items will not exist in standardized authority lists.
Despite these short comings, transforming EAD files into some sort of serialized RDF goes a long way towards publishing archival descriptions as linked data. This particular process is a good beginning and outputs valid information, just information that is not as linkable as possible. This process lends itself to iterative improvements, and outputting something is better than outputting nothing. But this particular proces is not for everybody. The archive whose content changes quickly, the archive with copious numbers of collections, or the archive wishing to publish the most complete linked data possible will probably not want to use EAD files as the root of their publishing system. Instead some sort of database application is probably the best solution.
In some ways MARC lends it self very well to being published via linked data, but in the long run it is not really a feasible data structure.
Converting MARC into serialized RDF through XSLT is at least a two step process. The first step is to convert MARC into MARCXML and then MARCXML into MODS. This can be done with any number of scripting languages and toolboxes. The second step is to use a stylesheet such as the one created by Stefano Mazzocchi to transform the MODS into RDF/XML — mods2rdf.xsl From there a person could save the resulting XML files on a Web server, enhance access via content negotiation, and called it linked data.
Unfortunately, this particular approach has a number of drawbacks. First and foremost, the MARC format had no place to denote URIs; MARC records are made up almost entirely of literals. Sure, URIs can be constructed from various control numbers, but things like authors, titles, subject headings, and added entries will most certainly be literals (“Mark Twain”, “Adventures of Huckleberry Finn”, “Bildungsroman”, or “Samuel Clemans”), not URIs. This issue can be overcome if the MARCXML were first converted into MODS and URIs were inserted into id or xlink attributes of bibliographic elements, but this is extra work. If an archive were to take this approach, then it would also behoove them to use MODS as their data structure of choice, not MARC. Continually converting from MARC to MARCXML to MODS would be expensive in terms of time. Moreover, with each new conversion the URIs from previous iterations would need to be re-created.
Encoded Archival Context for Corporate Bodies, Persons, and Families (EAC-CPF) goes a long way to implementing a named authority database that could be linked from archival descriptions. These XML files could easily be transformed into serialized RDF and therefore linked data. The resulting URIs could then be incorporated into archival descriptions making the descriptions richer and more complete. For example the FindAndConnect site in Australia uses EAC-CPF under the hood to disseminate information about people in its collection. Similarly, “SNAC aims to not only make the [EAC-CPF] records more easily discovered and accessed but also, and at the same time, build an unprecedented resource that provides access to the socio-historical contexts (which includes people, families, and corporate bodies) in which the records were created” More than a thousand EAC-CPF records are available from the RAMP project.
METS, MODS, OAI-PMH service providers, and perhaps more
If you have archival descriptions in either of the METS or MODS formats, then transforming them into RDF is as far away as your XSLT processor and a content negotiation implementation. As of this writing there do not seem to be any METS to RDF stylesheets, but there are a couple stylesheets for MODS. The biggest issue with these sorts of implementations are the URIs. It will be necessary for archivists to include URIs into as many MODS id or xlink attributes as possible. The same thing holds true for METS files except the id attribute is not designed to hold pointers to external sites.
Some archives and libraries use a content management system called ContentDM. Whether they know it or not, ContentDM comes complete with an OAI-PMH (Open Archives Initiative – Protocol for Metadata Harvesting) interface. This means you can send a REST-ful URL to ContentDM, and you will get back an XML stream of metadata describing digital objects. Some of the digital objects in ContentDM (or any other OAI-PMH service provider) may be something worth exposing as linked data, and this can easily be done with a system called oai2lod. It is a particular implementation of D2RQ, described below, and works quite well. Download application. Feed oai2lod the “home page” of the OAI-PMH service provider, and oai2load will publish the OAI-PMH metadata as linked open data. This is another quick & dirty way to get started with linked data.
Publishing linked data through XML transformation is functional but not optimal. Publishing linked data from a database comes closer to the ideal but requires a greater amount of technical computer infrastructure and expertise.
Databases — specifically, relational databases — are the current best practice for organizing data. As you may or may not know, relational databases are made up of many tables of data joined together with keys. For example, a book may be assigned a unique identifier. The book has many characteristics such as a title, number of pages, size, descriptive note, etc. Some of the characteristics are shared by other books, like authors and subjects. In a relational database these shared characteristics would be saved in additional tables, and they would be joined to a specific book through the use of unique identifiers (keys). Given this sort of data structure, reports can be created from the database describing its content. Similarly, queries can be applied against the database to uncover relationships that may not be apparent at first glance or buried in reports. The power of relational databases lies in the use of keys to make relationships between rows in one table and rows in other tables. The downside of relational databases as a data model is infinite variety of fields/table combinations making them difficult to share across the Web.
Not coincidently, relational database technology is very much the way linked data is expected to be implemented. In the linked data world, the subjects of triples are URIs (think database keys). Each URI is associated with one or more predicates (think the characteristics in the book example). Each triple then has an object, and these objects take the form of literals or other URIs. In the book example, the object could be “Adventures Of Huckleberry Finn” or a URI pointing to Mark Twain. The reports of relational databases are analogous to RDF serializations, and SQL (the relational database query language) is analogous to SPARQL, the query language of RDF triple stores. Because of the close similarity between well-designed relational databases and linked data principles, the publishing of linked data directly from relational databases makes whole lot of sense, but the process requires the combined time and skills of a number of different people: content specialists, database designers, and computer programmers. Consequently, the process of publishing linked data from relational databases may be optimal, but it is more expensive.
Thankfully, many archivists probably use some sort of behind the scenes database to manage their collections and create their finding aids. Moreover, archivists probably use one of three or four tools for this purpose: Archivist’s Toolkit, Archon, ArchivesSpace, or PastPerfect. Each of these systems have a relational database at their heart. Reports could be written against the underlying databases to generate serialized RDF and thus begin the process of publishing linked data. Doing this from scratch would be difficult, as well as inefficient because many people would be starting out with the same database structure but creating a multitude of varying outputs. Consequently, there are two alternatives. The first is to use a generic database application to RDF publishing platform called D2RQ. The second is for the community to join together and create a holistic RDF publishing system based on the database(s) used in archives.
D2RQ is a very powerful software system. It is supported, well-documented, executable on just about any computing platform, open source, focused, functional, and at the same time does not try to be all things to all people. Using D2RQ it is more than possible to quickly and easily publish a well-designed relational database as RDF. The process is relatively simple:
- download the software
- use a command-line utility to map the database structure to a configuration file
- edit the configuration file to meet your needs
- run the D2RQ server using the configuration file as input thus allowing people or RDF user-agents to search and browse the database using linked data principles
- alternatively, dump the contents of the database to an RDF serialization and ingest the result into your favorite RDF triple store
The downside of D2RQ is its generic nature. It will create an RDF ontology whose terms correspond to the names of database fields. These field names do not map to widely accepted ontologies & vocabularies and therefore will not interact well with communities outside the ones using a specific database structure. Still, the use of D2RQ is quick, easy, and accurate.
If you are going to be in Rome for only a few days, you will want to see the major sites, and you will want to adventure out & about a bit, but at the same time is will be a wise idea to follow the lead of somebody who has been there previously. Take the advise of these people. It is an efficient way to see some of the sights.