EssayCBM: Transparent Essay Grading with Rubric-Aligned Concept Bottleneck Models

This paper introduces EssayCBM, a novel approach to automated essay grading that prioritizes interpretability. By using a concept bottleneck, the system breaks down the grading process into evaluating specific writing concepts, making the evaluation process more transparent and understandable for both educators and students. The ability for instructors to adjust concept predictions and see the resulting grade change in real-time is a significant advantage, enabling human-in-the-loop evaluation. The fact that EssayCBM matches the performance of black-box models while providing actionable feedback is a compelling argument for its adoption. This research addresses a critical need for transparency in AI-driven educational tools.