In order to choose an algorithm to recommend student tutoring in the ASSISTments platform, a simulator was created based on historical ASSISTments data. Various contextual bandit models were tested in that simulator. It was found that the Disjoint and Hybrid LinUCB algorithms obtained the highest cumulative reward over a period of 100,000 trials. Consequently, Disjoint and Hybrid LinUCB will be implemented in the ASSISTments Reinforcement Learning Service to improve student learning outcomes.

• This report represents the work of one or more WPI undergraduate students submitted to the faculty as evidence of completion of a degree requirement. WPI routinely publishes these reports on its website without editorial or peer review.

Creator

• Healy, Alan

Publisher

• Worcester Polytechnic Institute

Identifier

• 53161
• E-project-032322-160420

Keyword

• Personalized Tutoring
• LinUCB
• Contextual Bandits
• Linear Thompson Sampling

Advisor

• Heffernan, Neil

Year

• 2022

Date created

• 2022-03-23

Resource type

• Major Qualifying Project

Major

• Computer Science

Rights statement

• In Copyright