Let us suppose that we’ve created a social approach to online learning, where-by our users not only take content from our Learning Environment, but actively add content to it as a part of their participation. One of the biggest problems facing those tasked with administrating such a platform is going to be information overload.
I admit this stage is somewhat ‘down the line’ in terms of a successful social learning environment, but to ignore planning for this would be short-sighted.
Simply leaving the task to administrators is not often a viable option; online learning is supposed to cut administration work, not make it worse. Developing a good taxonomy and naming convention can certainly help to spread the load, but this too has limitations.
What is required is a method of ‘Curating’ the content that your learners contribute. The role of the Curator is a vital one; sorting the wheat from the chaff and bringing some sort of order to what would otherwise be chaos.
Curators also go beyond these functions and use learning objects to tell us a story, providing deeper insight into what would otherwise be just a collection of ‘things’. But the job is a difficult one, requiring a subject matter expert and a good deal of time; see this post by Jeff Cobb on the need for good content curators.
In developing our new software, we’ve been looking into ways of “auto-curating” content which learners contribute to the learning environment. This is just one piece of a much larger puzzle that our latest product seeks to address, but it is a vital piece none-the-less.
A bit of background on how our learning environment works:
Learning objects are first of all organised into collections. Collections can be as be as big or as small as the editors of the learning environment deem appropriate. For example we may have a “dinosaur” collection, or perhaps we would look to do things at the level of the “T-rex”. Indeed, for those of us very involved in palaeontology we might choose to make our collections at an even ‘lower’ level, for example “the foot bones of the T-rex”.
Within collections sit objects. One object can exist in many collections. Objects can be any piece of digital information, from a web-link to a video, to an animation.
These objects carry with them an array of metadata, including details such as keywords. It is relatively easy to suggest that an object is like another object by using these details, but it is not a perfect match. And when we open our learning environment to contributions from all-comers, it is not easy to enforce a metadata tagging system which is always used, or always used correctly. Such data also fails to take into account the perceived quality of a learning object – do a lot of people view this object and rate it as a worthy object?
What if we could tell that an object was like another object without it actually sharing any metadata at all? We would be then in a position to automatically suggest which learning objects were related to each other and to start the Curation process without the need for human intervention.
Using a range of semantic web techniques, this is what we have attempted to do. Firstly, by adopting the Resource Descriptor Framework (RDF) in storing our learning objects, we are able to discover a lot more about the objects.
For example, think about Wines (I tend to veer towards alcohol metaphors when things get complex). The following statement breaks down the Stonleigh Sauvignon Blanc into an RDF readable format: (example taken from W3C)
SauvignonBlanc rdf:ID=”StonleighSauvignonBlanc”
locatedIn rdf:resource=”#NewZealandRegion”
hasMaker rdf:resource=”#Stonleigh”
hasSugar rdf:resource=”#Dry”
hasFlavor rdf:resource=”#Delicate”
hasBody rdf:resource=”#Medium”
SauvignonBlanc
Because of the way in which the information contained here is broken down, we can tell on a number of levels what a Stonleigh Sauvignon Blanc is like. It could be grouped with other Wines which are of Delicate flavour. Or perhaps we just want to group it with other Wines produced by the same Maker – Stonleigh. Or we can use combinations of multiple nodes to infer which wines the Stonleigh is most like.
Outside of RDF, we can also infer an amount of information about an object given other objects that we know connect to it in someway. Our software allows users to connect objects together as a part of their own “guides” – a way of knitting objects together to create a logical sequence of learning. Where these guides include some objects which share metadata, and some which do not, we are able to infer if an object is like another object.
Taking a crowd-sourced approach to grading our learning objects, we can also discover more about the usefulness of an object and its quality. This allows us to curate objects to not only find like objects, but also to find like objects of a certain quality.
In short, by utilising a number of semantic web techniques, we are aiming to create a learning environment that has the ability to organise any amount of content into suitable categories automatically. There remains a need for human intervention at some levels – for instance, the final “sense” check before things are sent live – but the workload is vastly reduced.
This is just one of the innovations we are looking to introduce with our new software, which we’ve aptly named Curatr. I’ll be blogging more on the features of Curatr in the coming weeks, but its safe to say we’re pretty excited about it.
