Search for: All records

This project examines the prospect of using AI-generated feedback as suggestions to expedite and enhance human instructors’ feedback provision. In particular, we focus on understanding the teaching assistants’ perspectives on the quality of AI-generated feedback and how they may or may not utilize AI feedback in their own workflows. We situate our work in a foundational college Economics class, which has frequent short essay assignments. We developed an LLM-powered feedback engine that generates feedback on students’ essays based on grading rubrics used by the teaching assistants (TAs). To ensure that TAs can meaningfully critique and engage with the AI feedback, we had them complete their regular grading jobs. For a randomly selected set of essays that they had graded, we used our feedback engine to generate feedback and displayed the feedback as in-text comments in a Word document. We then performed think-aloud studies with 5 TAs over 20 1-hour sessions to have them evaluate the AI feedback, contrast the AI feedback with their handwritten feedback, and share how they envision using the AI feedback if they were offered as suggestions. The study highlights the importance of providing detailed rubrics for AI to generate high-quality feedback for knowledge-intensive essays. TAs considered that using AI feedback as suggestions during their grading could expedite grading, enhance consistency, and improve overall feedback quality. We discuss the importance of decomposing the feedback generation task into steps and presenting intermediate results, in order for TAs to use the AI feedback. 

This project examines the prospect of using AI-generated feedback as suggestions to expedite and enhance human instructors’ feedback provision. In particular, we focus on understanding the teaching assistants’ perspectives on the quality of AI-generated feedback and how they may or may not utilize AI feedback in their own workflows. We situate our work in a foundational college Economics class, which has frequent short essay assignments. We developed an LLM-powered feedback engine that generates feedback on students’ essays based on grading rubrics used by the teaching assistants (TAs). To ensure that TAs can meaningfully critique and engage with the AI feedback, we had them complete their regular grading jobs. For a randomly selected set of essays that they had graded, we used our feedback engine to generate feedback and displayed the feedback as in-text comments in a Word document. We then performed think-aloud studies with 5 TAs over 20 1-hour sessions to have them evaluate the AI feedback, contrast the AI feedback with their handwritten feedback, and share how they envision using the AI feedback if they were offered as suggestions. The study highlights the importance of providing detailed rubrics for AI to generate high-quality feedback for knowledge-intensive essays. TAs considered that using AI feedback as suggestions during their grading could expedite grading, enhance consistency, and improve overall feedback quality. We discuss the importance of decomposing the feedback generation task into steps and presenting intermediate results, in order for TAs to use the AI feedback. 

Supporting novice programming learners at scale has become a necessity. Such a support generally consists of delivering automated feedback on what and why learners did incorrectly. Existing approaches cast the problem as automatically repairing learners’ incorrect programs; specifically, data-driven approaches assume there exists a correct program provided by other learner that can be extrapolated to repair an incorrect program. Unfortunately, their repair potential, i.e., their capability of providing feedback, is hindered by how they compare programs. In this paper, we propose a flexible program alignment based on program dependence graphs, which we enrich with semantic information extracted from the programs, i.e., operations and calls. Having a correct and an incorrect graphs, we exploit approximate graph alignment to find correspondences at the statement level between them. Each correspondence has a similarity attached to it that reflects the matching affinity between two statements based on topology (control and data flow information) and semantics (operations and calls). Repair suggestions are discovered based on this similarity. We evaluate our flexible approach with respect to rigid schemes over correct and incorrect programs belonging to nine real-world introductory programming assignments. We show that our flexible program alignment is feasible in practice, achieves better performance than rigid program comparisons, and is more resilient when limiting the number of available correct programs. 

The number of introductory programming learners is increasing worldwide. Delivering feedback to these learners is important to support their progress; however, traditional methods to deliver feedback do not scale to thousands of programs. We identify several opportunities to improve a recent data-driven technique to analyze individual program statements. These statements are grouped based on their semantic intent and usually differ on their actual implementation and syntax. The existing technique groups statements that are semantically close, and considers outliers those statements that reduce the cohesiveness of the clusters. Unfortunately, this approach leads to many statements to be considered outliers. We propose to reduce the number of outliers through a new clustering algorithm that processes vertices based on density. Our experiments over six real-world introductory programming assignments show that we are able to reduce the number of outliers and, therefore, increase the total coverage of the programs that are under evaluation.