We interviewed Phil Poekert of The University of Florida Lastinger Center on their upcoming data mining RFP.
Why Did You Announce the Data Mining RFP?
We announced the RFP inviting data mining within our platforms because we wanted to advance learning engineering research by increasing our network of collaborators.
Because we also do a significant amount of work in technology-enhanced professional development for early childhood and elementary educators toward improving kindergarten readiness and foundational literacy, we also began to see the possibilities of mining data from our Flamingo Learning platform to offer personalized professional development for teachers.
We’re big believers that significant breakthroughs in any field of human endeavor can be attributed to increasing and disseminating knowledge as well as advances in technology. By using large-scale learning platforms to offer personalized learning opportunities at scale and data science to enhance the efficacy of such interventions, we aim to do both.
The use of learning platforms to harvest data that can be useful in connecting learners with optimal resources and experiences that enhance the learning outcomes we’re trying to achieve.”
There’s a broad range of possibilities for what we might learn from the collaborations engendered by the data mining RFP, and we’re open to all of them. We don’t want to circumscribe the possible questions that researchers might ask of the data at the onset of this effort. However, we have pointed to a few questions that we’ve wondered about to catalyze a conversation about some possibilities. For example, as many as half of our users are accessing our professional development experiences in the Flamingo Learning platform using mobile devices. We’re wondering about the implications of mobile-friendly learning platforms for improving outcomes and for instructional design.
Our research into several years of Algebra Nation data has been incredibly insightful. A team of data scientists took a look at tens of thousands of student records. They developed a model that accounted for the variation in student performance among as many variables that we could take stock of student characteristics, teacher characteristics, school characteristics, district characteristics, as well as usage within the Algebra Nation platform.
We were delighted to learn that when we controlled for all the characteristics that we couldn’t change, there was still a statistically significant difference in student performance connected to their usage of the platform. Now, we’re using that statistical model as the basis of an algorithm that recommends learning content to students based upon an estimate of their current ability, and that estimate is updated as students interact with the platform further.
The biggest surprise has been the effort needed to appropriately instrument a learning platform to enable researchers to collect the required data for their experiments and guide experiments in personalized learning.
Significant time and energy have gone into creating valid measures of student engagement and ability and then developing the software needed to insert them into the platform appropriately. We imagine a time when the inertia needed to conduct experiments within learning platforms is reduced to facilitate more rapid, iterative research.