Large universities tend to spread their services across several departments to serve their substantial student base. It is very common for this to result in developing di erent systems, which end up in creating many disconnected data silos within the organization. Data isolation is one of the main bottlenecks that prevent unlocking the full potential behind exploiting such data, to provide a better experience at the level of application deployment and data analysis. The Open University is in the process of connecting their data silos by relying on the Linked Data principles within the LUCERO project. We discuss in this paper three use-cases through which we consume Linked Data produced at the Open University: (1) a student services use-case showing how we exploit data connections to deliver learning material related to courses through the university's main course information website; (2) a mobile course application that enables students to easily explore courses by subject, qualication or research topic; and (3) a Leanback TV service that provides students the ability to watch, with a degree of control, a set of podcasts grouped in di erent channels. Through these use cases, we highlight in this paper the advantages and e ects of consuming Linked Data within an organization.
Today's organizations generate a signi cant amount of information as a result of their daily operational activities. The Open University is a large higher and further education institution in the UK which is dedicated to distance learning. It delivers courses and quali cations to more than 200,000 students across the UK, as well as in other countries. To realise this, the OU's information repositories are among its core assets. However, as it is the case of most large organizations, di erent systems tend to be developed throughout the years to perform speci c tasks, which result in di erent disconnected data backbones. The absence of connections and links at the data level often makes data interchangeability between systems di cult and costly, which often leads to preventing the organization in fully exploiting their data.
Through the LUCERO project4, the OU has put in place a work ow to
connect their various data repositories following the Linked Data principles [
We aim in this paper to give an overview of the LUCERO project, and the work ow to generate Linked Data (Section 3). Then we discuss the consumption of Linked Data that we illustrate through three use-cases: (1) the student services use-case (Section 4), where we show how Linked Data helped discovering related learning material to courses, and displaying them on mainstream courses' web pages of the OU; (2) the mobile course browser (Section 5), through which we show how links between academic quali cations, subjects and courses helped delivering a better course browsing application to students; and (3) the Leanback TV webapp (Section 6) which provides a Leanback TV experience by relying on video material from the OU's Linked Data set. 2
In today's push towards creating semantics behind the data exposed over the web, we are witnessing an increase in the publishing side of Linked Data from educational bodies, following the original e ort realised at the OU7. For example, the University of Southampton have made their data available as Linked Data8. There are also plans from the University of Oxford to do the same9, as well as the University of Lincoln10, among others.
While more and more data is being exposed (as can be inferred from the
Linking Open Data cloud [
The LUCERO project aims to transform and expose institutional repositories within the Open University as Linked Data. We present in this section the work4 http://lucero-project.info 5 http://www.geonames.org 6 http://www.dbpedia.org 7 http://www.jisc.ac.uk/news/stories/2010/11/linkeddata.aspx 8 http://data.southampton.ac.uk/ 9 http://data.ox.ac.uk/ 10 http://data.lincoln.ac.uk/
ow involved in generating the OU Linked Data, in addition to the data already available. 3.1
As part of the LUCERO's main contributions, a work ow was put in place to handle the extraction and exposure of RDF elements from existing repositories. Figure 1 highlights the steps involved in this work ow, which we brie y describe in this section.
In the rst step, we Collect the information that resides in the original repositories across the various departments of the Open University. We set up di erent extractors to handle the extraction of the required elements from the sources. For example, the RSS extractor is customized to process video or audio Podcasts11 and publication data (i.e., Open Research Online12). The XML updater is used where the data is available in XML formal such as course descriptions. A scheduler handles an automatic daily process to check the changes that occurred at the data level, and run the work ow accordingly. 11 http://podcast.open.ac.uk 12 http://oro.open.ac.uk
Based on the identi ed items, we Extract the relevant RDF elements with the appropriate vocabulary to use. For example, in the case of a course item, we identify the need to model the course title, availability, cost, etc. In terms of the choice of vocabularies, our approach was to mostly reuse available ontologies, resulting in having in some cases more than one to apply. For example, AIISO13 and Courseware14 are both used, among others, to represent courses. An important component of the work ow is the Entity Name System, which ensures that each entity extracted gets a unique identi er across all our datasets, independently from its repository of origin. This guarantees for example that, when a book coming from the library catalogue dataset is part of course M36615 and a podcast is also related to course M366, both the book and podcast are pointing to the same course with a unique identi er. To achieve this, we design our code to apply a set of speci c patterns per entity type to follow prede ned URI pre xes (e.g. \http://data.open.ac.uk/course/" for Courses, or \http://data.open.ac.uk/library/" for library material), coupled with a unique identi er originating from the source.
Once the RDF data to add (or delete) are identi ed, we store them in a triple store, and expose them through a web portal accessible at http://data.open. ac.uk. This portal provides a SPARQL endpoint, as well as information about the datasets available, sample queries and applications. 3.2
The OU has already extracted and made available Linked Data from various
sources across the university. The process is continuous, with more data made
available whenever new resources are made available. The current set of data
that can be used include:
{ Courses Information: In addition to the ability of getting the
information through the OU website, courses information were also accessible in an
XML format, which made the transformation into Linked Data easier. This
dataset provides access to courses information including title, description,
availability by date and location, prices along with related books and other
material.
{ Podcasts: This dataset provides access to the OU podcast items, with their
corresponding title, related courses, subjects, video duration, etc. This set
has been extracted from the OU podcast RSS feeds16.
{ Library Catalogue: The OU library catalogue has been transformed from
MARC records [
This wealth of data has been exploited in various scenarios (see for
example [
Students are among the main consumers of data at the Open University. Providing them with a good experience in reaching the required information with the least e ort needed is key to the university's success. The rst use-case we present here is in the context of student services. Students are constantly looking for appropriate courses to enrol in at the Open University. Currently this is done through the main \Study at the OU" website24, where students can browse courses by subject, and get further information related to the courses and qualications they are counted towards. The Open University produces a substantial 17 http://www.youtube.com/user/TheOpenUniversity 18 http://openlearn.open.ac.uk 19 http://www.open.ac.uk/Arts/RED/index.html 20 http://www.openartsarchive.org 21 http://kmi.open.ac.uk/people/ 22 http://kmi.open.ac.uk/news/ 23 http://data.open.ac.uk/applications/ 24 http://www3.open.ac.uk/study amount of courses related material, for example through the OU YouTube Channel or podcast website. Such material can be of great bene t for students to form a better understanding of what to expect from a course, and helping them make a decision about what course to take. However the problem is that traditionally, explicit links between courses and related material are not available. Moreover, modifying the underlying data infrastructure (e.g., databases) or application to expose such links can be very costly in terms of development time and resources.
With the availability of the Linked Data portal at the Open University through data.open.ac.uk where items and their corresponding links are explicitly de ned, it became possible to easily get the entities related to speci c courses. Having courses, podcasts and OU YouTube material (among others) in a coherent representation following well de ned vocabularies, enabled a seamless extraction of needed entities through the available SPARQL endpoint. With this in place, we are in the process of extending the current courses' pages at the Open University, to o er students the functionality to explore learning material when visiting a course page. Figure 2 shows an example of related media, displayed on the right of the screen.
The formal \Study at the OU" site currently only describes Open University modules in text and yet there are other opportunities to demonstrate module content to prospective students through multimedia (podcast/iTunesU content and YouTube) or through the OU's Open Educational Resources project OpenLearn. Before the metadata associated with this content was represented as Linked Data the inherent links between the OU's OER, multimedia content and modules were not exploitable by machines. This meant that any attempt to supplement the OU's online prospectus with related content represented an unrealistic requirement in terms of man hours, especially considering that such content quickly becomes out-of-date.
The data used through this application are the Podcast and iTunesU datasets, in addition to the OpenLearn and YouTube data. Data is provided to this application through SPARQL queries returning all multimedia / OpenLearn content related to a module is made available as XML, and will be incorporated into the XML les that draw together data form other institutional OU sources before being rendered as HTML. 5
The Open University currently supports an iPhone and iPad application making content from its online prospectus available as an app. This application works well but there was a desire to improve it in two ways: (1) to make the application available across more platforms, and (2) to increase interactivity by making available material related to OU subject areas (e.g., Arts and Humanities, Maths and Computing, etc.) available on description pages as well as making more speci c related content available at module description level (e.g., content related more speci cally to \A101 - Introduction to Humanities"). Before LUCERO made these repositories available in a Linked Data format it was not possible for the OU to programmatically combine the data that creates the online prospectus with the content that it publishes in iTunesU, YouTube and OpenLearn. Although at a human level it was known that these resources related to modules (indeed frequently they are derived from module production process) this relationship was not expressed in a machine-readable way.
The newly-developed Study at the OU application is an HTML5 build made available in Android and iOS through thin client applications which serve to supplement a view onto the site through the devices browser (e.g., some of the navigation is handled through native application functionality whilst the content is a real-time web view). A similar approach is taken to the way the content will be surfaced via a Chrome add-on. Figure 3 shows a page through which students can explore courses and related content by quali cations, subjects, research, etc.
The data consumed in this app are from the Podcast, iTunesU, OpenLearn and YouTube datasets. The application reads (selected) data from the XML that is used to build the desktop online prospectus and combines this with data from data.open.ac.uk. When using the application a user can (whether they are at subject area level or module level) opt to view related resources. Podcast and YouTube content are handled through the devices native functionality and links to related OpenLearn content (webpages) are made available through the browser. The content read from data.open.ac.uk is dynamic and up-to-date so the application shows only the latest 3 resources from each area. This serves to keep the pages relevant and fresh and also maintains a good level of usability. 6
The OU has a plethora of video material spanning over various topics and subjects. With these hundreds of hours of interesting recordings, the question is: how do we engage the students in exploiting this vast resource of information, and exploring new areas of interests that might trigger new learning needs? With the current podcast portal, the student has to know what he/she is looking for in advance, before searching or going through the topics for the podcast of interest.
The purpose of Leanback TV applications is to provide viewers with a series of automatically selected set of video material that can be played one after the other, coupled with a degree of control over what to watch. We are witnessing an increase in Leanback TV services, provided for example by YouTube Leanback25, Redux TV26 and Vimeo Couchmode27. In these applications, a set of videos are preselected for viewers, with the ability to skip or choose among the set of videos, 25 http://www.youtube.com/leanback 26 http://redux.com/tv 27 http://vimeo.com/couchmode as well as choosing di erent channels. Such channels group the videos to watch into themes, such as YouTube's Trends and Best of YouTube.
In our context, we provide a Leanback TV experience over the set of learning podcast material. However, podcast information can be extracted from the current systems through its RSS feed, which doesn't provide the exibility we needed to manipulate and extract the data. We rely on the Linked Data provided through data.open.ac.uk, which we access through SPARQL queries to generate the set of videos to play through an HTML5 webapp interface. We used di erent queries to populate the channels with podcasts related to: OU research, OU learn and OU life. This is highlighted in Figure 4. Videos are also ltered based on their length, and only the ones that are below 10 minutes are shown to the users. Short videos have a higher chance to be watched by students, who know that something else will come up soon, in case such podcasts are of little interest to them. This will encourage students to explore new subjects, while relaxing in front of their TV or computer without having to choose what to watch next.
With the presence of SPARQL, it is possible to create a ne grained selection of videos. For example, one can create a channel about videos related to a speci c course, or a channel about the renaissance architecture. This is much harder to achieve through RSS feeds without further data customization at the source. Once the system is put in place, the options of creating new channels is endless, and very easy to extend. In the current demo28, the selected podcasts are mainly 28 http://labs.greenhughes.com/ldleanback/ { this application currently only works with the Google Chrome browser of general interest, where very specialized videos such as one hour lectures on mathematics are not needed. However, in other contexts, this could exactly be what's needed for a mathematics course webpage. Such di erent channels can now be easily created through just creating a di erent SPARQL channel. This is a very useful for scenarios where university teachers can have custom made channels for their classrooms. For example, this is the SPARQL query used to generate the OU Life podcast channel:
WHERE { ?podcast <http://digitalbazaar.com/media/depiction> ?thumb . ?podcast dc:title ?title . ?podcast wm:description ?desc . ?podcast <http://digitalbazaar.com/media/download> ?source . ?podcast wm:createDate ?createDate . ?podcast dc:subject <http://data.open.ac.uk/topic/ou_life> . ?podcast rdfs:type <http://data.open.ac.uk/podcast/ontology/
VideoPodcast> . } ORDER BY DESC(?createDate) LIMIT 25
Through this query, we are getting the thumbnail, description, title and source of the podcasts limited to the OU Life subject (i.e., < http : ==data:open:ac:uk= topic=ou lif e >), ordered by their creation date. Figure 5 displays a snippet of the results that can be processed through data.open.ac.uk. The application is based on Google TV Resources29, and the full implementation details are available online30. 7
Linked Data is transforming the way we access and consume data. From our three use-cases, we observe that the e ects on the organizations and the way they consume and provide their data can be at the following levels: Application development e ort: Linked Data reduces the cost involved to pull data out of repositories. In all the three use-cases, data was retrieved from one common endpoint using common tools and standards. Once the developer masters the way to extract Linked Data, and apply them in his/her 29 http://code.google.com/p/gtv-resources/ 30 http://www.greenhughes.com/content/html5-leanback-tv-webapp-brings-sparqlliving-room development environment, it will be very easy to apply the same methods in other applications. This will encourage developers to focus on the data usage, rather than on the technical di culties involved. In the case of the three applications described in this paper, the development has been performed without having to go back to the original data maintainers. This, by itself, is considered a substantial decrease in the cost involved within the organization.
User experience: We observe that Linked Data enabled the development of applications that will enhance user experience. This observation surely needs to be backed by some kind of survey once the applications are fully deployed and accessed by a substantial amount of users. However we foresee that, by for example simply aiding the users by decreasing the required e ort in nding relevant material while looking for courses, or interesting podcasts, our applications will be positively embraced by our end-users.
Connecting to others: In this paper, we have mostly focused on the bene t of connecting data silos internal to the organization. However, we also provide links to external datasets, including for example to the countries in which courses are available in GeoNames, to the description of the Open University in http://education.data.gov.uk and to the postcode units of OU buildings in http://data.ordnancesurvey.co.uk. Providing such external links not only supports the enrichment of our own data, but also provides users with additional entry points into it, which are not dictated by the university's view on the provided information. We are for example investigating scenarios in which resources at the OU could be linked to TV programs from the BBC, providing ways to discover study material relating to the topics treated in a TV program.
In this paper, we have presented three use-cases that demonstrate how Linked Data can be consumed in a large organization such as the Open University. The goal was, through these examples, to discuss the bene ts, issues and e ects of integrating Linked Data practices in such an organization. Indeed, based on our experience in building the Linked Data platform of the OU and supporting other departments in consuming such data, we believe that the success of Linked Data will not be based on the development of \killer apps", but on solving in a cost e ective way numerous problems that organizations, data managers and developers are facing. We also wanted to emphasis the importance of organizational aspects in the adoption of Linked Data in the context of a large organization. We selected the three presented use cases as they represent clear examples where Linked Data is consumed in projects led by non-academic services of the University (communications/online services, student services and IT), being clearly driven by the \business needs" of the organization, rather than by the application of a particular technological solution.
As we mentioned at the beginning of the paper, many other universities are
now following the same approach as we have developed within the LUCERO
project, to expose public university resources as Linked Data. While this trend
is still at an early stage, interesting cases and ways of consuming linked data are
expected to emerge from it. Being able to discover educational resources related
to a topic or a course independently from the way and place in which they
were produced seems an obvious scenario. It however generates new challenges
related to the aggregation of data from di erent institutions31, as well as to the
availability of common classi cation schemes for these resources [