Skip to main content Skip to navigation

Exercise Hierarchical Feature Enhanced Knowledge Tracing

Project Overview

The document explores the application of generative AI in education, focusing on a novel framework for knowledge tracing within intelligent education systems. This framework enhances traditional methods by integrating hierarchical features derived from exercise text, such as knowledge distribution, semantic characteristics, and the difficulty level of exercises. By leveraging these features, the framework aims to improve the accuracy of predicting student performance. Experimental results indicate that this innovative approach significantly outperforms existing models, underscoring the critical role of hierarchical features in achieving more precise knowledge tracing. Overall, the findings suggest that incorporating generative AI into educational systems can lead to enhanced learning outcomes by providing tailored insights into student understanding and performance.

Key Applications

Exercise Hierarchical Feature Enhanced Knowledge Tracing (EHFKT)

Context: Online education systems, targeting students who interact with educational exercises.

Implementation: The framework uses embedding from the Bert model to extract features and employs RNNs to model student responses. It integrates knowledge distribution, semantic features, and difficulty of exercises.

Outcomes: Reported performance improvements in knowledge tracing accuracy, better representation of questions, and precise predictions of student performance.

Challenges: Challenges include the unpredictability of exercise difficulty and the complexity of interpreting semantic clusters.

Implementation Barriers

Technical

Existing knowledge tracing models have limitations in considering various attributes of exercises.

Proposed Solutions: The proposed EHFKT framework addresses these limitations by incorporating hierarchical features and sophisticated modeling techniques.

Data Availability

Lack of open datasets with exercise records that include text information.

Proposed Solutions: The authors created a custom experimental dataset from a real-world online education system.

Project Team

Hanshuang Tong

Researcher

Yun Zhou

Researcher

Zhen Wang

Researcher

Contact Information

For information about the paper, please contact the authors.

Authors: Hanshuang Tong, Yun Zhou, Zhen Wang

Source Publication: View Original PaperLink opens in a new window

Project Contact: Dr. Jianhua Yang

LLM Model Version: gpt-4o-mini-2024-07-18

Analysis Provider: Openai

Let us know you agree to cookies