Up to Design Issues
Government data is being put online to increase accountability, contribute valuable information about the world, and to enable government, the country, and the world to function more efficiently. All of these purposes are served by putting the information on the Web as Linked Data. Start with the "low-hanging fruit". Whatever else, the raw data should be made available as soon as possible. Preferably, it should be put up as Linked Data. As a third priority, it should be linked to other sources. As a lower priority, nice user interfaces should be made to it -- if interested communities outside government have not already done it. The Linked Data technology, unlike any other technology, allows any data communication to be composed of many mixed vocabularies. Each vocabulary is from a community, be it international, national, state or local; or specific to an industry sector. This optimizes the usual trade-off between the expense and difficulty of getting wide agreement, and the practicality of working in a smaller community. Effort toward interoperability can be spent where most needed, making the evolution with time smoother and more productive.
This, 2009, is the year for putting government data online. Both US and UK governments made public commitments toward open data. The TED talk on Linked Data was in February. Groups from the Guardian to the Sunlight Foundation had already been pushing for it for a long time. People like Watchdog.net, mysociety.org, and govtrack.us had been pushing by publishing government data themselves in various formats, including Linked Data.
So if you want to do this, what should you do? This article addresses this question very briefly, and makes a set of points which will probably be outdated by later developments, but answer a set of relevant question, asked or not.
Government data is put online typically for 3 reasons:
Each of these purposes is best served by using Linked Data techniques.
In general Linked Data is:
Open: Linked Data is accessible through an unlimited variety of applications and applications because it is expressed in open, non-proprietary formats.
Modular: Linked Data can be combined (mashed-up) with any other piece of Linked Data. For example, government data on health care expenditures for a given geographical area can be combined with other data about the characteristics of the population of that region in order to assess effectiveness of the government programs. No advance planning is required to integrate these data sources as long as they both use Linked Data standards.
Scalable: It's easy to add more Linked Data to what's already there, even when the terms and definitions that are used change over time.
The essential message is that whatever data format people want the data in, and whatever format they give it to you in, you use the RDF model as the interconnection bus. That's because RDF connects better than any other model.
That's enough about why it is useful. That is elaborated elsewhere, but it can be difficult for those familiar with other technologies to understand the difference. Sometimes it is better just to do it.
The chances are quite high that the data your department/agency runs off will be largely in relational databases, often with a large amount in spreadsheets.
There are two philosophies to putting data on the web. The top-down one is to make a corporate or national plan, by getting committees together of all the interested parties, and make a consistent set of terms (ontology) into which everything fits. This in fact takes so long it is often never finished, and anyway does not in fact get corporate or national consensus in the end. The other method experience recommends is to do it bottom up. A top-level mandate is extremely valuable, but grass-roots action is essential. Put the data up where it is: join it together later.
A wise and cautious step is to make a thorough inventory of all the data you have, and figure out which dataset is going to be most cost-effective to put up as linked data. However, the survey may take longer than just doing it. So, take some data.
A really important rule when considering which data could be put on the web is not to threaten or disturb the systems and the people who currently are responsible for that data. It often takes years of negotiation to put together a given set of data. The people involved may be very invested in it. There are social as well as technical systems which have been set up. So you leave the existing system undisturbed, and find a way of extracting the data from it using existing export or conversion facilities. You add, a thin shim to adapt the existing system to the standard.
Ok, so you have some data. What form is it in?
There are (2009) a number of open source tools for putting relational databases up as Linked Data, D2RServer and Triplify being two.
These each use a mapping file, in some language, to explain how the database structure actually represents things and the relationships. 1
You probably don't want to to run a publicly available server on your existing database unless it is generally set up for high volume use. You might want to take a copy of the whole database, and run a live semantic web server from it, or you can generate the RDF once and make a copy of that to serve.
It is wise and friendly and interoperable, when you public RDF data, to use terms other people are already sharing. Like foaf:name for the name of a person, or dc:title for the title of something, and so one. Like geo:lat and geo:long for latitude and longitude2. There are a number of these, growing of course. The Semantic Web Interest Group is a community which can help you find them: there are also online tools such as Swoogle, Sindice, etc.
In many organizations a surprising amount of information, sometimes critical information, is emailed around in spreadsheets. Much of the early recovery.gov data was published in spreadsheet form. Some of these are raw tables, with a header in the top row. These are close to raw data. You can export them as a comma-separated (or tab-separated) file, CSV. Others are spreadsheets with a lot of substructure, and little headings and notes all over them for the human user. These are less easy to convert.
There are a number of tools for converting the format of a spreadsheet, typically in CSV form, into RDF.
If you have existing data in XML, first, put that XML up on the web while you think. Then, figure out what the XML is about, what things and what relationships. Then, commission or write a program, possibly a simple script, maybe written in XSLT, or your favorite scripting language, to convert each XML file into RDF. You might need to add a file which points to all the things you have data about, if they are not already linked.
Ok, so your data is not in any of the above forms. It is in a proprietary format, or managed by a proprietary program. But there is some way you can get at it. So someone will have to write a program somewhere, to get it out, and convert it to one of the Linked Data standard forms.
(It is actually fairly simple. First, you think of what things the data is about. You make up URIs for those things. Suppose for example your data is about books and shelves. You decide the URI for the books will be http://id.example.com/id/isbn/123457890 and the URIs for shelves will be like http://id.example.com/id/shelf/746 . Then you write a (CGI) script, which, when given that a URI like that extracts the data about the book (including which shelf it is on) and outputs it, or similarly for the shelf (including a list of the books on the shelf). It outputs it in RDF/XML or N3. That script is your web server of virtual linked data.)
If you have an existing web site with, maybe, a page about each thing, there is an easy way of putting the data in those pages into Linked Data. You can change the scripts which generate the site so that the data which is behind each page is in fact put into the page so that it can be re-extracted by others as data. The technology to do this is called RDFa 3. An alternative is for the each web page to have a parallel page which has the data in RDF/XML. 4
Ok, so you have your data in RDF as Linked Data. Now what?
The semantic web toolkit includes the SPARQL query language which allows a client anywhere on the net to query a SPARQL service. Some methods of publishing data, like D2RServer, provide a built-in SPARQL service. If you have generated a bunch of linked data, then there are various products, free or commercial, which will scoop it up into a "triple store" and provide a SPARQL service.
A SPARQL service is a generally useful tool for technically aware users. Many clients and analytical tools just use a SPARQL server. A SPARQL server looks for patterns in the data and for each match, or outputs what it found in one of a number of formats, including constructed RDF, XML and, in some cases, JSON, and maybe even CSV.
SPARQL, then, can be used as an RDF to XML converter. You amass a heap of linked data. Then you think of a combination of data, involving connections across different data. There is a SPARQL query for that data with the results expressed in XML. That SPARQL query can be encoded into a long URI, a URI for a virtual XML document for that particular view.
Some SPARQL servers also support JSON as an output format. This is easy to use in Web Applications.
The priority first is to get raw data onto the net, and preferably converted into Linked Data form. This is partly because there may be other sites, commercial or not, who pick it up and make great interfaces to that data. Of course there are times when the government site must provide a easy human interface for ordinary users to access the data.
There are many routes to pretty HTML for real users. Tools like Exhibit provide facetted browser views, given a configuration set up by the web master, for example.
Note, though, there are two ways though that a department or agency web site can never be expected to compete with external sites. One is because there are as yet no user interface techniques which allow a normal user to create their own query, (though tools like Tabulator are getting close).
The second is that an external site will add value to the data by joining it to other data from different sites for a particular purpose. If the Department of Transport publishes road accident data, a cycling site selects the cycle accident subset, and can publish it as a map adding cycle routes and hills, and cycle shops. An agency publishes data about the amount of money given to different towns, another maps it against the per capital income levels in those towns. And so on in uncountable permutation.
An informal random sample of some public feedback suggests that there are users who would prefer each of these formats above, so a system which generates them automatically is clearly called for.
When you write or generate a small RDF file for each dataset exported, the results can be harvested as more useful linked data to form a catalog. Like the data, this can be distributed form as linked data, and also sucked into a repository to be indexed and SPARQLed. Remember that, as with the data, RDF allows you to mix vocabularies, so you can record everything you or others may feel is important about the datasets. This provenance information is very valuable. It clearly is one of the many areas this note touches on which much more could be said.
Neither does it really address licensing issues. In the US, government data is generally in the Public Domain. It is good to put the fact that a given resource has a given license in a machine-readable way. The creative commons cc:license term is appropriate. Creative commons also have produced a "CC0" waiver which disclaims all rights appropriately (and where possible) for each country.
A very common and important concern is the privacy of data which contains personally identifiable nformation. This article does not suggest that all data should be made public, nor does it discuss issues with anonymisation of data. Systems where PIP is an issue will probably not be an early choice when selecting those to put on the web. However, in cases in which these issues have already been resolved and the data is already public but not in the standard form, converting it to Linked Data is an excellent idea. In general, new government systems should be built to be aware of the provenance of the data they use, and of the appropriate use to which it may be put. But the design of these accountable systems is another topic we do not have space for here.
This brief note is too short to go into great detail, and has ignored many important topics. It has stressed the practical technical steps. Deeper information, about techniques and also about the social issues and challenges, are being produced frequently elsewhere. Many cities have Semantic Web gatherings or meetup groups, which can be a source of mutual support for those involved in or interested in the technology. The W3C eGov Interest Group is an international group of people sharing challenges and solutions.
 D2RServer will generate a default mapping file, which will not make a very good RDF graph. Browsing the resulting RDF with am RDF browser (such as Tabulator) will however often show up the deficiencies and suggest improvements
 WGS84 latitude and longitude, like you get from a normal GPS unit. (more)
 RDFa is used, for example, in the UK Civil Service Jobs web site. (example)
 Separate RDF/XML web pages are used, for example, in the BBC programmes data. Here content negotiation gives RDF/XML to data clients, and HTML to document browsers. (example)
Thanks for input to this article from Nigel Shadbolt and Danny Weitzner. Thanks also to the chairs (John Sheridan and Kevin Novak) and members of the W3C eGov interest group, and all those in UK and US governments with whom we have discussed these issues at these early stages.
Up to Design Issues