Challenges of Modeling Longitudinal Development

Does synthetic xAPI data hold promise for all sorts of new learning research?

Within a research context, there is a need for large and relevant activity data sets to test hypotheses regarding competency assertion. But where do these data sets come from?

Whether on account of the cost and logistics involved in producing longitudinal data sets, or due to the nature of privacy concerns in the space of research data, it can be daunting to actually find or produce the data sets necessary to advance research in the space of learning.

A novel solution, mimicking efforts in the space of artificial intelligence research, is to create large scale synthetic data sets. The data represented in the sets can be modeled to serve as an appropriate facsimile of real-world data, are easier to produce, and sidestep any issues related to personally identifiable information.

With this in mind, our research team on the STEEL-R project modeled activity within xAPI Profiles and used this to generate synthetic data sets that could represent learning activity over time — learning activity that then could be mined in much the same way that it could have been were it collected from real-life learners.

In STEEL-R, we have hot-rodded the Generalized Intelligent Framework for Tutoring (GIFT) with an xAPI Profile designed to send patterns of behavior about learners who take part in training scenarios downstream where the data can be used as evidence by the math model of the CaSS competency and skills system — which in turn automates the assertion of competency attainment.

Needing to test the system, we turned to DATASIM — a capability that renders synthetic xAPI data based on whatever xAPI Profile patterns and simulation specifications are fed to it.

Read the paper — including what we found we were able to do… and what we had challenges doing… over at ERIC.