Free response evaluation via neural network for an IMathAS system

Abstract

A fully interactive class with mixed reality and simulation learning should provide many free response types for students to learn beyond numerical answers and multiple choice. Essay and string responses in the IMathAS homework system have to be manually graded, making the free response questions difficult to generate instant feedback. The ability to write questions with automatic feedback during active lecture offer improvements to the current systems and provide an opportunity for critical thinking to occur. The following study provides framework for an interpretive neural network to be implemented into any IMathAS system. These responses can be in the form of equations, words and sentences, or pictures. Findings show that correctly trained networks using manually graded artifacts can be more than 90% accurate in providing feedback to a correct answer in student practice, allowing for lessons that guide students towards correct and well-phrased answers using their own words, and can even assign partial credit. The findings imply that Marzano's taxonomy level of analysis can be reached using the IMathAS system and that critical thinking methods can be directly applied for scoring. When integrated into the existing system, simulation-based or mixed reality homework can have free responses and the grades can be transferred via learning tool interoperability connection into the institutional learning management system for direct scoring in the gradebook.