Integrated Data-driven Technologies for Individualized Instruction in STEM Learning Environments

This project aims to develop intelligent learning technology designed to react to individual student performance data, so as to personalize instruction. Such technology has significant potential to transform the American educational system by providing a low-cost way to adapt learning environments to individual students' needs and by informing advanced research on human learning. This project will create the technology for a new generation of data-driven Intelligent Tutoring Systems (ITSs), enabling the rapid creation of individualized instruction that supports learning in science, technology, engineering, and mathematics (STEM). The net result of this work will be a modular framework of educational data mining methods that offer student-adaptive, individualized support at multiple granularities, that have been implemented, iteratively refined, and empirically validated for learning impact and robustness across systems. This project will develop hierarchical data-driven, interpretable, and robust models that optimize student learning. Moreover, it will investigate whether integrating hierarchical data-driven agent decision-making with user-initiated decisions can help students learn to make better decisions for their learning. Teaching students to make effective decisions can fundamentally transform educational assessment: the emphasis should not be just on what students have learned, but on whether students can learn and adapt in productive ways in future situations. By providing individualized instruction using data, it has the potential to make individualized learning support accessible to a broad audience, including students that are traditionally underrepresented in STEM fields. These efforts serve the national interests by strengthening the nation's ability to develop and diversify the STEM workforce.

The goal of this project is to develop and empirically evaluate a general hierarchical data-driven framework that would induce hierarchical hints and adaptive hierarchical pedagogical decision making policies across three STEM domains, including logic, probability, and programming, where building traditional ITSs is extremely challenging. More specifically, this project will 1) advance research on data-driven approaches to ITSs by adapting them to make subgoal hints and hierarchical decisions similar to those of human experts; 2) evaluate the robustness of our general hierarchical data-driven framework by comparing it to flat data-driven approaches not only on each individual ITS but also across ITSs; and 3) close the loop by using data-driven policies to improve students' decision-making and their long-term problem-solving abilities. The proposed work is poised to have a significant impact by making ITSs more effective, by improving student performance in STEM domains, and by teaching students to make effective pedagogical decisions. If successful, it will close the loop by using data-driven policies to support student decision-making and eventually improve their long-term problem-solving abilities through hybrid human-machine interactive decision-making in-vivo experimentation.