Purpose: Research demonstrates that learners who receive one-on-one tutoring perform better than those who receive only traditional classroom instruction and tools that systematically address students' misconceptions and errors whilethey are working on problems – not after they complete an assignment or exam – enable those students to learn better and faster. In this project, the team will fully develop a web-based artificial intelligence tutor to provide personalized support students as they solve math word problems.

Project Activities: During Phase I in 2022,the project team developed a prototype of a Natural Language Processing and Machine Learning model that provides instantaneous feedback to students as they solve math word problems, including hints prior to a student setting up a word problem.  At the end of Phase I, a pilot study, with four educators and 34 high school students, demonstrated that the prototype functioned as planned, that students indicated they received just-in-time hints and supports while completing word problems, and educators stated that the information generated by the dashboard was useful and appropriate for providing individualized instruction for each student.

In Phase II of the project, the team will fully develop the product, including adding a voice input as alternative to text entry based, by continuing to add data to optimize the Natural Language Processing and Machine Learning models that setup problems and provides hints using the specific language of the word problem, fully developing the user interface for students and educators, and developing the reporting dashboard framework for web-hosted reports. After development concludes, a pilot study will test the feasibility and usability, fidelity of implementation, and the promise of the product for improving science learning. The team will collect data from 60 grade 4 and 5 classes with approximately 15 students per class, with half randomly assigned to use the product and the other half to use business-as-usual activities for the same course content.  Researchers will compare pre-and-post scores using the Stanford Achievement Test and the Mathematics Problem Solvingassessment at pre-test, post-test, and follow-up. Researchers will gather cost information using the "ingredients method" and will include all expenditures on things such as personnel, facilities, equipment, materials, and training.

Product: With previous ED/IES SBIR awards the company developed the StepWise Virtual Tutor, which uses Artificial Intelligence to personalize tutoring and assessment for Algebra. The product facilitates students in exploring many valid pathways to a solution and the system tracks each step in the work, catches algebraic and arithmetic errors, and provides hints and presents information to educators.  This project is developing the StepWise Virtual Tutor for Math Word Problems to apply AI techniques to review students' verbal statements about a chosen approach to a word problem presented. Natural Language Processing (NLP) techniques will also be employed to analyze the text of a math word problem for clues about its solution. A dashboard will provide information about each students' progress. In addition, the product will guide students in breaking down word problems using the Self-Regulated Strategy Development (SRSD), a model of solving problems in writing that involves goal-setting, self-monitoring, self-reinforcement, and self-instruction.

Related IES Projects: StepWise Virtual Tutor for Algebra I (EDIES16C0011)
StepWise Virtual Tutor for Common Core Algebra I (EDIES15C0015)
StepWise Virtual Tutor for Math Word Problems Using SRSD (91990022C0036)

Video Demonstration of the Phase I Prototype: https://www.youtube.com/watch?v=VNsl2svgcFc

Company Website: http://www.querium.com