An Empirical Study of Finding Similar Exercises
Project Overview
The document highlights the transformative role of generative AI in education, particularly through the innovative approach to the Finding Similar Exercises (FSE) problem aimed at improving intelligent test paper generation and personalized learning. By introducing the pre-trained Chinese language model BERT Edu and a multi-task model named ExerciseBERT, the authors address various challenges including data scarcity, labeling noise, and the semantic comprehension of educational exercises. These models significantly enhance the accuracy of exercise similarity matching, leading to improved educational outcomes. The findings underscore the potential of generative AI to facilitate personalized learning experiences, thereby reshaping assessment and instructional methods in educational settings.
Key Applications
ExerciseBERT - a pre-trained model for finding similar exercises
Context: Junior mathematics education, targeting teachers and students
Implementation: Developed a pre-trained language model (BERT Edu) and utilized exercise normalization and multi-task learning to improve exercise matching.
Outcomes: Improved performance in identifying similar exercises, leading to personalized learning opportunities and enhanced test paper generation.
Challenges: Data scarcity, high labeling noise, and insufficient understanding of exercise semantics.
Implementation Barriers
Data-related barrier
Scarcity of labeled educational data and high costs associated with expert labeling.
Proposed Solutions: Utilized a pre-trained model to leverage unlabeled data and implemented confidence learning to manage label noise.
Technical barrier
Diversity in exercise representations leading to difficulties in matching.
Proposed Solutions: Introduced exercise normalization to standardize different forms of exercises for better comparison.
Understanding barrier
Insufficient semantic understanding of exercises complicating the similarity assessments.
Proposed Solutions: Adopted multi-task learning to enhance logical and semantic representations of exercises.
Project Team
Tongwen Huang
Researcher
Xihua Li
Researcher
Contact Information
For information about the paper, please contact the authors.
Authors: Tongwen Huang, Xihua Li
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