Research
The Problem
There is a growing focus on and importance of end-of-year examinations for both students and teachers with the No Child Left Behind Act (NCLB) and its successor, Every Student Succeeds Act (ESSA) (Every Student Succeeds Act, 2015; No Child Left Behind Act, 2002). This has led many states to develop their own standardized curricula and end-of-year assessments. One such state, Virginia, utilizes a framework called Standards of Learning (SOLs), developed by the Virginia Department of Education (VDOE). Under this framework, each grade level and subject’s yearly instruction is guided by SOLs that are aligned with the end-of-year test students must take in most grade levels. While student performance on these exams is one critical way school quality is evaluated in Virginia, there is a severe lack of questions from which students can practice due to the state having a limited test bank. Specifically, the VDOE is only able to provide roughly 55 questions per grade level for the most recent SOL end-of-year test available. Clearly, there is a need to provide additional questions from which to practice to ensure success on these exams. Additionally, standardized tests are currently not built to activate student interest, which is associated with better performance on learning activities and tests (Bernacki & Walkington, 2018; Walkington, 2013). By giving students the chance to practice questions similar to standardized test questions in a low-stakes environment, activating student interest, therefore, could mitigate the effects of stereotype threat and the resulting lower performance it causes on these assessments for students from minority backgrounds (Spencer et al., 2016). With the proliferation of Large Language Models (LLMs), it is now conceivable to generate an infinite test bank of math questions aligned with student interests, which is the goal of this project.
The Solution: MATHWELL
MATHWELL, or Mathematics Assessment through Word Problems Extracted from Large Language Models, is a LLM designed to create a limitless test bank of questions for students to practice with that are aligned with their interests. Using MATHWELL, students and teachers will be able to generate unique math word problems based on mathematical operations and topics such as soccer, basketball, superheroes, etc. The goal is to create a system that can not only generate good math word problems, but also accurately solve the problems it creates in order to give students feedback on their answers. To this end, we will train the model to generate step-by-step solutions to the problems it creates in order to give students detailed feedback.
How You Will Help
Your help will be invaluable in creating MATHWELL. You will help classify questions so that the model can learn to create solvable, appropriate questions and step-by-step solutions that are aligned with specific topics.