Mathematical solution process step grading and feedback generation method based on deep learning

CN121836992BActive Publication Date: 2026-05-29SOUTHWEST JIAOTONG UNIV
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN Β· China
Patent Type
Patents(China)
Current Assignee / Owner
SOUTHWEST JIAOTONG UNIV
Filing Date
2026-03-13
Publication Date
2026-05-29

AI Technical Summary

Technical Problem

Existing technologies are prone to semantic short circuits in mathematical problem-solving, failing to effectively model the rigorous logical hierarchy between mathematical solution steps, leading to misjudgments in grading and feedback generation.

Method used

We employ a deep learning-based approach, enhancing step context encoding through a pre-trained mathematical language understanding model and a temporal convolutional network. We then construct a hierarchical step logic graph by combining a multi-head attention mechanism and a causal discovery algorithm. Finally, we use a hierarchical graph neural network for reasoning to generate a high-dimensional step state vector sequence, enabling precise step-level grading and feedback.

Benefits of technology

It effectively avoids semantic short-circuit interference, improves the reliability and interpretability of the step logic diagram, enhances the credibility of error diagnosis, and generates more accurate step-level corrections and feedback.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121836992B_ABST
    Figure CN121836992B_ABST
Patent Text Reader

Abstract

The application discloses a mathematical solution process step grading and feedback generation method based on deep learning, and belongs to the technical field of auxiliary teaching, which comprises the following steps: obtaining student answer texts and dividing the texts into step sequences; generating step semantic encoding; enhancing context information through a time sequence convolution network; the core lies in constructing a hierarchical and anti-semantic short circuit step logical relationship graph: first, pre-labeling logical roles and levels for the steps, embedding a learnable attenuation factor based on hierarchical differences in the attention mechanism to constrain correlation calculation, then using causal discovery to verify and eliminate false correlation edges that violate logical timing and independence, and finally obtaining a high-dimensional vector representing the deep representation of the step logical state through hierarchical graph neural network reasoning; based on this, step-level error detection and classification are carried out, and targeted feedback is generated. The application effectively suppresses the semantic short circuit connection between steps, improves the accuracy and reliability of logical relationship modeling, and thus realizes more accurate automatic step-level grading and feedback.
Need to check novelty before this filing date? Find Prior Art