The evolution of Learning Analytics is changing what we can see of learning; it is also changing how we see learning and it is beginning to change the nature of learning itself and giving rise to new pedagogies. Given appropriate tools and starter data, we believe learners can gather and use data themselves to build personalised feedback models. With support, they can use these models to aid goal setting and tracking and provide opportunities for metacognition and self-reflection. Learners are often seen as passive objects of learning analytics (LA) initiatives and the opaque, black-box nature of many of LA systems reinforces this tendency. This research explores ways to involve students as active participants in shaping LA to their needs and encouraging them to be critical, data literate users of their own learning data.
We have developed proof-of-concept, interactive, white-box probabilistic graphical models which students can interact with visually, and which are designed to give a sense of agency and an interconnected understanding of their learning activities. We are using qualitative methods to try to learn what other data nodes would be impactful for students’ understanding and that would prompt and support metacognitive and self-reflection activities. These models are being developed further, tested and validated. Finally in future phases, we will conduct quantitative and qualitative analyses of their impact on student learning and learners.
There are many modelling approaches (Chrysafiadi & Virvou, 2013) but not all are accessible to the student themselves and not all lend themselves to effective reasoning approaches. Probabilistic Graphical Modelling is a branch of machine learning that studies how to use probability distributions to describe the world and to make useful predictions about it, combining Bayesian probability and graph theory. Bayesian Networks are simple constructs in some ways but have been proven to be a powerful tool in student modelling (Millán, Loboda, & Pérez-de-la-Cruz, 2010).
In this research, course data has been gathered with students in a 1st year ‘Internet of Things’ module and has been parsed into interactive Bayesian Networks. These are being shared with the next cohort of students and a qualitative study will be completed to learn alongside students which models they find useful, and more importantly, what other data models would be perceived as helpful and insightful for them as they evolve as learners. New models from this qualitative work with students will be built and presented back to students for another iteration of quantitative and qualitative analysis.
We completed work on collating and discretising this data and building initial Bayesian Networks models in Sep/Oct 2017 - these are currently being validated. In future iterations of the research, students will be able to interact with course variables and see expected correlations with course performance, based on the performance of a past cohort. Students will be able to condition these variables on their own actual or expected performance levels. Eventually, either directly or indirectly, they will be able to add new nodes that they believe are important to their educational experience. While these current proof-of-concept models are an interesting starting point, we believe their real impact will only become fully apparent when we start to model on nodes that emerge from our qualitative work with students on identifying their own learning data priorities.
Topics: Learning Analytics: Research and Practice , Topics: Students as Partners
