Human-Assisted Computation for Auto-Grading

Abstract

In this paper, we present a novel auto-grading framework that can automatically grade student assignments without prior knowledge of the answers. The idea is crowd-sourcing or human-assisted computation that extract knowledge from a large number of people to make predictions using hypothesis testing and Bayesian analysis. We also explore the possibilities of combining this framework with an educational chatbot software interface (e.g., the Facebook Messenger chatbot platform), in order to utilize the built-in image annotation feature that facilitates the assignment submission process in large classes.