Auxiliary Task Guided Interactive Attention Model for Question Difficulty Prediction
Project Overview
The document explores the implementation of generative AI in education, focusing on a method called QDiff, which predicts the difficulty levels of academic questions using Bloom's Taxonomy. It emphasizes the importance of personalized learning experiences in online education, advocating for adaptive question difficulty based on individual learner performance. The QDiff approach employs a multi-task learning framework with an interactive attention mechanism, resulting in improved accuracy for predicting both the complexity of questions and their corresponding Bloom's taxonomy classifications. The findings indicate that this method significantly enhances traditional techniques, thereby potentially transforming online learning environments by fostering tailored educational experiences that cater to the diverse needs of students. Overall, the use of generative AI, exemplified by QDiff, represents a promising advancement in creating more effective and engaging educational tools that align with personalized learning goals.
Key Applications
QDiff
Context: Application in online learning platforms for students in grades VI to XII in a K-12 education system.
Implementation: The method uses a transformer-based model (like BERT) and an interactive attention mechanism to predict the difficulty and Bloom's taxonomy levels of questions.
Outcomes: Improved accuracy in predicting question difficulty and enhanced personalized learning experiences.
Challenges: The method's effectiveness may be reduced when Bloom's labels are not available for certain datasets.
Implementation Barriers
Data availability
Bloom's taxonomy labels may not always be provided for some datasets, which is essential for the proposed QDiff method.
Proposed Solutions: The study proposes soft-labeling methods to generate Bloom's taxonomy levels from datasets that only have difficulty labels.
Project Team
Venktesh V
Researcher
Md. Shad Akhtar
Researcher
Mukesh Mohania
Researcher
Vikram Goyal
Researcher
Contact Information
For information about the paper, please contact the authors.
Authors: Venktesh V, Md. Shad Akhtar, Mukesh Mohania, Vikram Goyal
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