Computer-based learning environments are a treasure trove of data. But while webshops are good at translating user interactions into product recommendations, using such data to support a student in learning complex tasks is anything but a no-brainer. Manuel Valle Torre, PhD student at the Leiden-Delft-Erasmus Centre for Education and Learning (LDE-CEL), wants to change that.
It takes more than an advanced algorithm to analyse a bunch of data and turn it into learner support
‘There is a disconnect between research and what is happening in actual institutions,’ Manuel Valle Torre says. ‘In LDE-CEL, we are really researching and promoting Learning Analytics – the use of data of learners and the learning environment to understand and optimise learning. But in the universities, if there are any implementations, they are small-scale and isolated.’ Part of the reason is that teachers have zero time available for adopting solutions that are not plug-and-play. More importantly, especially when it comes to learning complex tasks, it takes more than an advanced algorithm to analyse a bunch of data and turn it into wisdom. ‘Different learners require a different kind of support. So do different teaching styles. There is a very large human aspect to Learning Analytics.’
The human aspect
It is exactly this human aspect that led Manuel to LDE-CEL. ‘I started off in business informatics, but the objective there was always to sell more or to reduce cost. Pretty much the only course I really enjoyed was the one on business intelligence, which is close to data science.’ He subsequently decided to leave Mexico and pursue a master’s in computer science at TU Delft to increase his technical prowess in data engineering.
Working as a research engineer, he drifted into Learning Analytics. ‘I found this even way more exciting than just data, as it was about people. Now, pursuing my PhD, I have found my sweet spot.’ Having already supported several researchers, he knows his way around at LDE-CEL. Around the Netherlands as well, given the fact that he owns four bikes and a bike trailer. With COVID restrictions easing up, he again bikes to work. ‘I do try to cluster all my meetings in a single day each week, to have a few days of pure focus.’
Learning Analytics is not just about data, it is about humans
Real-time learner feedback
In his research, Manuel focusses on when students are learning complex skills, such as differential equations or learning to program. ‘You can’t draw an exact set of instructions for learning these skills, the learning process is not a straight path from A to B,’ he says. ‘I want to use learning analytics to identify their learning progress and then provide feedback when they need it.’
The common way to apply learning analytics is to calculate features from ‘flattened’ data –the number of clicks by a student, the number of problems they tried to solve, the number of incorrect answers they gave. Manuel, on the other hand, will look at sequences of data traces to generate a learner model, taking into account the time aspect. It makes a difference if a student, for example, alternates correct answers with incorrect ones, or if he first provides ten incorrect answers, followed by ten correct ones. ‘Suppose a student appears stuck, I do not yet know what a proper intervention will look like, but I do want it to be real-time. Latest by the end of their learning action.’
Data sequences can reveal more about a learner than ‘flattened’ data
And then there is the student
To make things even more complicated: what may be a good intervention according to scientific articles on educational engineering, may not be what the student is looking for. Manuel: ‘For a long time, learning optimisation used to be about success rates only – passing or failing, dropout rates. But it has become more about learning, about the student’s objectives. Some students who take an online course really want to master all the topics, others just want to pass, a third may want to focus on only a single module.’
His colleague Gabrielle Martins investigates such goal setting. It is up to Manuel to take this into account. ‘We have interviews ongoing to find out the feedback students want and are willing to consider – information they want to set their goals prior to starting a course, information during their course for time management and monitoring how well they are doing, and information upon completion to reflect on their learning process.’
Validation
Manuel has pretty much wrapped up the literature review from which he will now select the most promising methods and trends. He will then validate these by means of experiments, ideally in a setting with lots of data already available. ‘I may choose to collaborate with someone in PRIME, the blended learning approach applied by TU Delft for teaching mathematics,’ he says. ‘Or I may plug my support system into a MOOC on programming.’
(Almost) plug and play
By the end of his PhD, Manuel would like to have developed a decision framework that allows teachers to implement some kind of learner support into their program. ‘It probably won’t be completely plug and play, but it would be really cool if at least the somewhat more scientifically skilled teachers can say: “I’m doing this type of course and I need this kind of support. According to this framework I need to do this and this.”’
I want to develop a decision framework allowing teachers to implement some kind of learner support into their program.
Although his research is aimed at a higher educational setting, he expects his results may end up having a much wider application. ‘I think it can be transferred to teaching block-based programming to primary and secondary school students. It’s a syntax-free programming language that allows them to do highly complex tasks, controlling Lego robots for example.’ One way or another, humans will be central to his solutions. ‘Machine learning may be essential for finding patterns in the data. But a person is then required to interpret those patterns into what it means for students.’