Next up at LAK 13 Leuven I’m joining a full paper session entitled ‘Interpreting Data Mining Results with Linked Data for Learning Analytics: Motivation, Case Study and Directions’ by Mathieu D’Aquin and Nicolas Jay.
All papers can be found in the following drop box
Mathieu is that rare thing, a data scientist.
This is the subject of a panel discussion later today and if we believe the hype of the conference organising panel, it’s going to be an important and in demand role in society an one that is currently lacking capacity.
Mathieu starts us off with two great slides which I repeat below. The first is a naive view of learning analytics. The second includes the really interesting bit ‘Interpretation’.
This is going to slide into the uber technical I think, but I’ll first echo my own point earlier today when I hinted that analytics is all well and good but it’s only of use if;
a) the course is sound
b) the instruction is sound
c) the student are engaged (and engaged enough to look at the analytics and action the insights)
d) the student and instructor have the appropriate data literacy skills to make sound interpretations
Anyway, on with the linked data view.
And here’s a view of the web of linked data as of 2011 (apparently it’s now grown too complex to represent in this manner.
You can’t see the detail but it has a bubble for University data and one for research publications among media providers such as BBC etc(!)
Mathieu next showcased the OU open data project
data.open.ac.uk hosts datasets obtained from public data repositories at the Open University and applications making use of these data. Currently, the datasets relate the publications, courses and Audio/Video material produced at the Open University, as well as the people involved in making them. All these data are available through standard formats (RDF and SPARQL) and are (in most cases) available under an open license (Creative Commons Attribution 3.0 Unported License).
The OU has a Course Profile Facebook Application that entices student to share the courses they have, are and intend to complete with friends. This allows them to represent each students’ trajectory as a sequence of courses. From 8806 sequences (students) they obtained 126 sequential patterns occurring in at least 100 sequences (course A —> course E —> course M etc). The need for background information becomes apparent. We need the context of the courses themselves in order to draw inferences from the patterns. Once we have that we can build relations that apply to the items of patterns eventually providing a hierarchy of patterns. Are you with me so far?
Next we are able to explore the hierarchy and realise these benefits
My own conclusions are a little lacking here! We have patterns of courses as student trajectory, we know what those courses are, we have other course related information. I presume it allows us to predict successful pathways. Hey ho.
So the session was about the importance of linked (big) data. I’d draw attention to a session I blogged a month ago on Analytics and Institutional Capabilities where we heard from Ranjit Sidhu (Sid) (@rssidhu) that often big data projects = big waste of money. Sid counselled to look to local data first. This LAK session seems to have done both. It’s local data but enhanced by the Facebook App.