Design and Application of a Homework Behavior Recognition and Personalized Feedback System

Abstract

Existing automatic grading systems primarily focus on the correctness of answers, often overlooking the behavioral features demonstrated by students during the homework process. To address this issue, this paper proposes and implements a student homework behavior analysis and personalized feedback system based on homework images. The system uses image processing and multidimensional feature modeling to extract six key features, including handwriting clarity, correction frequency, and post-correction accuracy. It also combines classification and clustering methods to identify typical behavior patterns and generate feedback. Unlike previous studies that focus only on result-based assessments, this system constructs an end-to-end closed-loop from homework collection and behavior modeling to feedback output, incorporating an interpretability mechanism to enhance teacher trust. Experimental results on real homework data show that the system significantly outperforms baseline methods in terms of behavior recognition accuracy (87.6%), F1 score (0.842), and teacher satisfaction (4.63/5). The study not only validates the effectiveness of the proposed method but also provides valuable insights into the exploration of personalization and interpretability in intelligent education systems.