A Focus on Student Education: Determining Student Attitudes Towards Transparent Autograding and Developing Artificially Intelligent Tools to Help Students Succeed

This thesis is composed of two parts both relating to helping students succeed. First, the focus is on determining how we can help students feel more comfortable using an AI tool that can provide them immediate feedback. Second, machine learning algorithms are explored in relation to tracking student tasks to encourage healthy study habits. The development of effective autograders is key for scaling assessment and feedback. While AI based autograding systems for open-ended response questions have been found to be beneficial for providing immediate feedback, autograders are not always liked, understood, or trusted by students. Our research tested the effect of informing students about key aspects of the autograder on their attitudes towards those autograders. Providing more transparent information about the autograders increased students' perceptions of how effective the autograder was at grading their work, but did not improve other related attitudes—such as willingness to be graded by them on a test—relative to the control group which was not given extra information about the autograder. However, this lack of impact may be due to higher measured student trust towards autograders in this study than in prior work in the field. We briefly discuss possible reasons for this trend. The second chapter of this thesis addresses the issue of students struggling to maintain focus on tasks. This chapter's research explains the development process of a tool aimed to help teach students to focus by providing them with scaffolds so they can practice focusing. Prior research has used keyboard and mouse data to track user affective states and intentions, but task tracking software primarily has used application information for accurate classification. Using the Behacom dataset, this paper explores the training of various machine learning models on keyboard, mouse, and system data. Through feature selection and training machine learning models, this research is successful in developing a model to accurately classify user application usage. This information can then be used to determine if a user is on or off task in a way that maximizes user privacy.