Fast Learning from Unlabeled Episodes for Next-Generation Tailoring (FLUENT)

The Challenge

The TLA has great potential to improve learning experiences for individual students through data interoperability across the DoD. To achieve this potential, a new capability is needed to collect, manage, and share details about student learning interactions across multiple systems and contexts. Access to such data would allow the DoD to tailor education and training in response to the evolving needs of the student.

The Solution

An adaptive recommendation system that uses learner performance data from xAPI statements within a Learning Record Store (LRS) to continuously replan and prioritize the sequence of learning activities presented to the student in order to optimize learning opportunities.

About the Project

The purpose of this project was to create a system, FLUENT, that interprets xAPI data generated by learning activities to make recommendations for additional learning activities that advance the student’s learning goals. Learning activities are sequenced together into Learning Episodes that encapsulate information about each learning event in the sequence, its effectiveness for different types of users, and the context of each learning event (e.g., the string of learning events that preceded it). For example, FLUENT observes a current learning episode representing a type of failure by the student on a particular learning module, which is recorded in the LRS. From this episode, FLUENT has the ability to call up closely matching previous episodes in which the learner made learning choices to overcome similar failures. FLUENT interprets these past episode outcomes in the context of the current episode and generates a recommendation for the learner.

The ability of FLUENT to make these recommendations is driven by meta-adaptation, which is used to discover interactions among tools and environmental factors from detailed LRS event data streams. The FLUENT technical approach to meta-adaptation enhances the LRS and Learner Profile by recognizing interactions and coordinating across tools. The FLUENT recommender can also use xAPI statements to track and measure performance against the requirements for obtaining proficiency across a range of competencies.

This initial research created prototype algorithms and software to demonstrate a proof of the concept that was tested and evaluated using simulated data. This resulted in the development of meta-adaptation data requirements and design documentation, exemplar data models and APIs, an API design guide, open-source software, and prototype systems. In April 2017, a baseline evaluation of the FLUENT architecture was undertaken with human participants at Fort Bragg, North Carolina, which reflected lessons learned during previous pilot testing. This empirical test included 30 participants acting as learners over a four-day period. One of the key outcomes of the Fort Bragg event was capturing recorded data that tracked learner activity and system performance to be used for analysis in the design-based research process. The last phase of the project supported the testing and evaluation of relevant technical specifications and standards for the TLA, which helped develop the TLA Data Strategy for 2019.

Publications