Tag Archives: course description

HECoS, a new subject coding system for Higher Education

You may have missed that just before Christmas HECoS (the Higher Education Classification of Subjects) was announced. I worked a little on the project that lead up to this, along with colleagues in Cetis (who lead the project), Alan Paull Serices and Gill Ferrell, so I am especially pleased to see it come to fruition. I believe that as a flexible classification scheme built on semantic web / linked data principles it is a significant contribution to how we share data in HE.

HECoS was commissioned as part of the Higher Education Data & Information Improvement Programme (HEDIIP) in order to find a replacement to JACS, the subject coding scheme currently used in UK HE when information from different institutions needs to be classified by subject. When I was first approached by Gill Ferrell while she was working on a preliminary study of to determine if it needed changing, my initial response was that something which was much more in tune with semantic web principles would be very welcome (see the second part of this post that I wrote back in 2013). HECoS has been designed from the outset to be semantic web friendly. Also, one of the issues identified by the initial study was that aggregation of subjects was politically sensitive. For starters, the level of funding can depend on whether a subject is, for example, a STEM subject or not; but there are also factors of how universities as institutions are organised into departments/faculties/schools and how academics identify with disciplines. These lead to unnecessary difficulties in subject classification of courses: it is easy enough to decide whether a course is about ‘actuarial science’ but deciding whether ‘actuarial science’ should be grouped under ‘business studies’ or ‘mathematics’ is strongly context dependent. One of the decisions taken in designing HECoS was to separate the politics of how to aggregate subjects from the descriptions of those subjects and their more general relationships to each other. This is in marked contrast to JACS where the aggregation was baked into the very identifiers used. That is not to say that aggregation hierarchies aren’t important or won’t exist: they are, and they will, indeed there is already one for the purpose of displaying subjects for navigation, but they will be created through a governance process that can consider the politics involved separately from describing the subjects. This should make the subject classification terms more widely usable, allowing institutions and agencies who use it to build hierarchies for presentation and analysis that meet their own needs if these are different from those represented by the process responsible for the standard hierarchy. A more widely used classification scheme will have benefits for the information improvement envisaged by HEDIIP.

The next phase of HECoS will be about implementation and adoption, for example the creation of the governance processes detailed in the reports, moving HECoS up to proper 5-star linked data, help with migration from JACS to HECoS and so on. There’s a useful summary report on the HEDIIP site, and a spreadsheet of the coding system itself. There’s also still the development version Cetis used for consultation, which better represents its semantic webbiness but is non-definitive and temporary.

A short project on linking course data

Alasdair Gray and I have had Anna Grant working with us for the last 12 weeks on an Equate Scotland Technology Placement project looking at how we can represent course information as linked data. As I wrote at the beginning of the project, for me this was of interest in relation to work on the use of schema.org to describe courses; for the department as a whole it relates to how we can access course-related information internally to view information such as the articulation of related learning outcomes from courses at different stages of a programme, and how data could be published and linked to datasets from other organisations such as accrediting bodies or funders. We avoided any student-related data such as enrolments and grades. The objectives for Anna’s work were ambitious: survey existing HE open data and ontologies in use; design an ontology that we can use; develop an interface we can use to create and publish our course data. Anna made great progress on all three fronts. Most of what follows is lifted from her report.

(Aside: at HW we run 4-year programmes in computer science which are composed of courses; I know many other institution run 3/4-year courses which are comprised of modules. Talking more generally, course is usefully ambiguous to cover both levels of granularity; programme and module seem unambiguous.)

A few Universities have already embarked on a similar projects, notably the Open University, Oxford University and Southampton University in the UK, and Muenster and the American University of Beirut elsewhere. Southampton was one of the first Universities to take the open linked data approached and as such they developed their own bespoke ontology. Oxford has predominantly used the XCRI ontology (see below for information on the standard education ontologies mentioned here) to represent data, additionally they have used MLO, dcterms, skos and a few resource types that they have defined in their own ontology. The Open University has the richest data available, the approach they took was to use many ontologies. Muenster developed the TEACH ontology, and the American University of Beirut used the CourseWare and AIISO ontologies.

The ontologies reviewed were: AIISO, Teach, CourseWare, XCRI, MLO, ECIM and CEDS. A live working draft of the summary / review for these is available for comment as a Google Doc.

Aiiso (Academic Institution Internal Structure Ontology) is an excellent ontology for what it is designed for but as it says, it aims to describe the structure of an institution and doesn’t offer a huge amount in the way of particular properties of a course. Teach is a better fit in terms of having the kind of properties that we wished to use to describe a course, however doesn’t give any kind of representation of the provider of the course. CourseWare is a simple ontology with only four classes and many properties with with Course as the domain, the trouble with this ontology is that it is closely related to the Aktors ontology which is no longer defined anywhere online.

XCRI and MLO are designed for the advertising of courses and as such they miss out some of the features of a course that would be represented in internal course descriptions such as assessment method and learning outcomes.  Neither of these ontologies show the difference between a programme and a module. ECIM is an extension of MLO which provides a common format for representing credits awarded for completion of a learning opportunity.

CEDS (Common Education Data Standards) is an American ontology which provides a shared vocabulary for educational data from preschool right up to adult education.  The benefits of which are, that data can be compared and exchanged in a consistent way.  It has data domains for assessment, learning standards, learning resources, authentication and authorisation.  Additionally it provides domains for different stages of education e.g. post-secondary education. CEDS is ambitious in that it represents all levels of education and as such is a very complex and detailed ontology.

XCRI, MLO (+ ECIM) and CEDS can be grouped together in that they differentiate between a course specification and a course instance, offering or section.  The specification being the parts of a course that remain consistent from one presentation to the next, whereas the instance defines those aspects of a course that vary between presentations for example location or start date. The advantage of this is that there will be a smaller amount of data that will require updating between years/offerings.

An initial draft of a Heriot Watt schema applying all the ontologies available was made. It was a mess, however it became apparent the MLO was the predominant ontology.  So we chose to use MLO where possible and then use other ontologies where required.  This iteration  resulted in a course instance becoming both an MLO learning opportunity instance and a TEACH course in order to be able to use all the properties required.  Even using this mix of ontologies we still needed to mint our own terms.  This approach was a bit complex and TEACH does not seem to be widely used, we therefore decided to use MLO alone and extend it to fit our data in a similar way that already started by ECIM.

The final draft is shown below. Key:  Green= MLO, Purple=MLO extension, Blue=ECIM / previous alteration to MLO Yellow= generic ontologies such as Dublin core and SKOS.  In brief, we used subtypes of MLO Learning Opportunities to describe both programmes and modules. The distinction between information that is at the course specification level and that which is at the course instance level was made on the basis of whether changing the information required committee approval. So things that can be changed year on year without approval such as location, course leader and other teaching staff are associated with course instance; things that are more stable and require approval such as syllabus, learning outcomes, assessment methods are at course specification level.

mloExtension2

We also created some instance data for Computer Science courses at Heriot-Watt. For this we use Semantic MediaWiki (with the Semantic bundle). Semantic forms were used for inputting course information, the input from the forms is then shown as a wiki page. Categories in mediawiki are akin to classes, properties are used to link one page to another and also to relate the subject of the page to its associated literals. An input form has the properties inbuilt such that each field in the form has a property related to it. Essentially the item described by the form will become the object in the stored triple, the property associated with a field within the form will form the predicate of the stored triple and the input to the field will form the subject of a triple. A field can be set such that multiple values can be entered if separated by commas, and in this case a triple will be formed for each value.  I think there is a useful piece of work that could be done comparing various tools for creating linked data (e.g. Semantic MediaWiki, Calimachus, Marmotta) and evaluating whether other approaches (e.g. WordPress extensions) may improve on them. If you know of anything along such lines please let me know.

We have little more work to do in representing the ontology in Protege and creating more instance data, watch this space for updates and a more detailed description than the image above. We would also like to evaluate the ontology more fully against potential use cases and against other institutions data.

Anna has finished her work here now and returns to Edinburgh Napier University to finish her Master’s project. Alasdair and I think she has done a really impressive job, not least considering she had no previous experience with RDF and semantic technologies. We’ve also found her a pleasure to work with and would like to thank her for her efforts on this project.

Linked learning highlights from Florence #LILE2015

davidI was lucky enough to go to Florence for some of the pre WWW2015 conference workshops because Stefan Dietze invited me to talk at the Linked Learning workshop “Learning & Education with the Web of Data“. Rather than summarize everything I saw I would like to give brief pointers to three presentations from that workshop and the “Web-base Education Technologies” (WebET 2015) workshop that followed it that were personal highlights. Many thanks to Stefan for organizing the conference (and also to the Spanish company Gnoss for sponsoring it).

Semantic TinCan

I’ve followed the work on Tin Can / xAPI / Experience API since its inception. Lorna and I put a section about it into our Cetis Briefing on Activity Data and Paradata, so I was especially interested in Tom De Nies‘s presentation on TinCan2PROV: Exposing Interoperable Provenance of Learning Processes Through Experience API Logs. Tin Can statements are basically elaborations of “I did this,” providing more information about the who, how and what referred to by those three words. Tom has a background in provenance metadata and saw the parallel between those statements and the recording of actions by agents more generally, and specifically with the model behind the W3C PROV ontology for recording information about entities, activities, and people involved in producing a piece of data or thing. Showing that TinCan can be mapped to W3C PROV is of more than academic interest: the TinCan spec provides only one binding, JSON but the first step of Tom’s work was to upgrade that to JSON-LD and then through the mapping to PROV allow any of the serializations for PROV to be used (RDF/XML, N3, JSON-LD), and to bring the Tin Can statements into a format that allows them to be used by semantic technologies. Tom is hopeful that the mapping is lossless, you can try it yourself at tincan2prov.org.

Linking Courses

I also have an increasing interest in the semantic representation of courses, there’s some interest in adding Courses to schema.org, but also within my own department some of us would like to explore advantages of moving away from course descriptors as PDF documents to something that could be a little more connected with each other and the outside world. Fouad Zablith’s presentation on Interconnecting and Enriching Higher Education Programs using Linked Data was like seeing the end point of that second line of interest. The data model Fouad uses combines a model of a course with information about the concepts taught and the learning materials used to teach them. Course information is recorded using Semantic MediaWiki to produce both human readable and linked data representations of the courses across a program of study. A bookmarklet allows information about resources that are useful for these courses to be added to the graph–but importantly attached via the concept that is studied, and so available to students of any course that teaches that concept. Finally, on the topic of several courses teaching the same concepts, sometimes such repetition is deliberate, but sometimes it is unwanted.

Showing which concepts (middle) from one course (left) occur in other courses (right). Image from Fouad Zablith's presentation (see link in text)
Showing which concepts (middle) from one course (left) occur in other courses (right). Image from Fouad Zablith’s presentation (see link in text)

Fouad  showed how analysis of the concept – course part of the graph could be useful it surfacing cases of where there were perhaps too many concepts in a course that had been previously covered elsewhere (see image, above)

Linking Resources

One view of a course (that makes especial sense to anyone who thinks about etymology) is that it is a learning pathway, and one view of a pathway is as an Directed Acyclic Graph, i.e. an ordered route through a series of resources.  In the WebET workshop,  Andrea Zielinski presented A Case Study on the Use of Semantic Web Technologies for Learner Guidance which modelled such a learning pathway as a directed graph and represented this graph using OWL 2 DL.  The conclusion to her paper says “The approach is applicable in the Semantic Web context, where distributed resources often are already annotated according to metadata standards and various open-source domain and user ontologies exist. Using the reasoning framework, they can be sequenced in a meaningful and user-adaptive way.” But at this stage the focus is on showing that the expressivity of OWL 2 DL is enough to represent a learning pathway and testing the efficiency of querying such graphs.

On Semantics and the Joint Academic Coding System

Lorna and I recently contributed a study on possible reforms to JACS, a study which is part of a larger piece of work on Redesigning the HE data landscape. JACS, the Joint Academic Coding System, is mainatained by HESA (the Higher Education Statistics Agency) and UCAS (Universities and Colleges Admissions Service) as a means of classifying UK University courses by subject; it is also used by a number of other organisations for classification of other resources, for example teaching and learning resources. The study to which we were contributing our thoughts had already identified a problem with different people using JACS in different ways, which prompted the first part of this post. We were keen to promote technical changes to the way that JACS is managed that would make it easier for other people to use (and incidentally might help solve some of the problems in further developing JACS for use by HESA and UCAS), which are outline in the second part.

There’s nothing new here, I’m posting these thoughts here just so that they don’t get completely lost.

Subjects and disciplines in JACS

One of the issue identified with the use of JACS is that “although ostensibly believing themselves to be using a single system of classification, stakeholders are actually applying JACS for a variety of different purposes” including Universities who “often try to align JACS codes to their cost centres rather than adopting a strictly subject-based approach”. The cost centres in question are academic schools or departments, which are discipline based. This is problematic to the use of JACS to monitor which subjects are being learnt since the same subject may be taught in several departments. A good example of this is statistics, which is taught across many fields from Mathematics through to social sciences, but there are many other examples: languages taught in mediaeval studies and business translation courses, elements of computing taught in electronic engineering and computer science and so on. One approach would be to ignore the discipline dimension, to say the subject is the same regardless of the different disciplinary slants taken, that is to say statistics taught to mathematicians is the same as statistics taught to physicists is the same as statistics taught to social sciences. This may be true at a very superficial level, but obviously the relevance of theoretical versus practical elements will vary between those disciplines, as will the nature of the data to be analysed (typically a physicist will design an experiment to control each variable independently so as not to deal with multivariate data, this is not often possible in social sciences and so multivariate analysis is far more important). When it comes to teaching and learning resources something aimed at one discipline is likely to contain examples or use an approach not suited to others.

Perhaps more important is that academics identify with a discipline as being more than a collection of subjects being taught. It encapsulates a way of thinking, a framework for deciding on which problems are worth studying and a way of approaching these problems. A discipline is a community, and an academic who has grown up in a community will likely have acquired that community’s view of the subjects important to it. This should be taken into account when designing a coding scheme that is to be used by academics since any perception that the topic they teach is being placed under someone else’s discipline will be resisted as misrepresenting what is actually being taught, indeed as a threat to the integrity of the discipline.

More objectively, the case for different disciplinary slants on a problem space being important is demonstrated by the importance of multidisciplinary approaches to solving many problems. Both the reductionist approach of physics and the holistic approach of humanities and social sciences have their strengths, and it would be a shame if the distinction were lost.

The ideal coding scheme would be able to represent both the subject learnt and the discipline context in which it was learnt.

JACS and 5* data

Tim Berners-Lee suggested a 5 star deployment scheme for open data on the world wide web:
* make your stuff available on the Web (whatever format) under an open licence
** make it available as structured data (e.g., Excel instead of image scan of a table)
*** use non-proprietary formats (e.g., CSV instead of Excel)
**** use URIs to denote things, so that people can point at your stuff
**** link your data to other data to provide context

Currently JACS fails to meet the open licence requirement for 1-star data explicitly, but that seems to be a simple omission of a licensing statement that shows the intention that JACS should be freely available for others to use. It is important that this is fixed, but aside from this, JACS operates at about 3-star level. Assigning URIs to JACS subjects and providing useful information when someone accesses these URIs will allow JACS to be part of the web of linked open data. The benefits of linking data over the web include:

  • The identifiers are globally unique and unambiguous, they can be used in any system without fear of conflicting with other identifiers.
  • The subjects can be referenced globally by humans by from websites, emails, and by computer systems in/from data feeds and web applications.
  • The subjects can be used for semantic web approaches to representing ontologies, such as RDF.
  • These allow relationships such as subject hierarchies and relationships with other concepts (e.g. academic discipline) to be represented independently of the coding scheme used. An example of this is SKOS, see below.

In practical terms, implementing this would mean:

  • Devising a URI scheme. This could be as simple as adding the JACS codes to a suitable base URI. For example H713 could become http://id.jacs.ac.uk/H713
  • Setting up a web service to provide suitable information. Anyone connecting to that URL would be redirected to information that matched parameters in their request. A simple web browser would request an HTML page and so be redirected to http://id.jacs.ac.uk/H713.html; web applications would request data in a machine readable form such as xml, rdf or json.

The main overhead is in setting up, maintaining and managing the data provided by the web service, but Southampton University have already set one up for their own use. (The only problem with the Southampton service–and I believe Oxford have done something similar–is a lack of authority, i.e. it isn’t clear to other users whether the data from this service is authoritative, up to date, used under a correct license, sustainable.)

JACS and SKOS

SKOS (Simple Knowledge Organization System) is a semantic web application of RDF which provides a model for expressing the basic structure and content of concept schemes such as thesauri, classification schemes, subject heading lists, taxonomies, folksonomies, and other similar types of controlled vocabulary. It allows for the description of a concept and the expression of the relationship betweens pairs of concepts. But first the concept must be identified as such, with a URI. For example:
jacs:H713 rdf:type skos:concept
In this example jacs: is shorthand for the JACS base URI, http://id.jacs.ac.uk/ as suggested above; rdf: and skos: are shorthand for the base URIs for RDF and SKOS. This triple says “The thing identified by http://id.jacs.ac.uk/H713 is a resource of type (as defined by RDF) concept (as defined by SKOS)”.

Other assertions can be made about the resource, e.g. the preferred label to be used for it and a scope note for it.
jacs:H713 skos:prefLabel “Production Processes”
jacs:H713 skos:scopeNote “The study of the principles of engineering as they apply to efficient application of production-line technology.”

Assuming the other JACS codes have been suitably identified, relationships between them can be described:
jacs:H713 skos:broader jacs:H710
jacs:H713 skos:related jacs:H830

Once JACS is on the semantic web relationships between the JACS subjects and things in other schemas can also be represented
http://example.org/123 dct:subject jacs:H713
(The resource identified by the URI http://example.org/123 is about the subject identified by jacs:H713).