Multi-Task Learning based Online Dialogic Instruction Detection with Pre-trained Language Models
Project Overview
This document explores the application of generative AI in education through the development of a multi-task learning framework designed to detect dialogic instructions in online classes using pre-trained language models. By focusing on identifying different types of instructional dialogue, the system provides valuable feedback to instructors, which enhances teaching effectiveness and promotes student engagement and understanding. The framework employs a contrastive loss mechanism to improve classification accuracy and incorporates strategies to mitigate errors from automatic speech recognition (ASR) systems. Findings reveal that this approach significantly outperforms baseline models in recognizing effective teaching strategies, suggesting that generative AI can play a pivotal role in transforming online education by fostering better communication and interaction between instructors and students.
Key Applications
Multi-task learning framework for dialogic instruction detection
Context: Online education, targeted at instructors and educators
Implementation: Utilizes pre-trained language models and a contrastive loss approach for instruction detection from online class videos
Outcomes: Improved detection accuracy for dialogic instructions, enhanced instructor feedback, and better online teaching practices
Challenges: Variability in teaching styles, errors from ASR transcriptions, and the need for robust detection methods
Implementation Barriers
Technical Barrier
Inconsistent quality of dialogic instructions and variability in pedagogical styles among instructors complicate detection accuracy.
Proposed Solutions: Implementing a multi-task learning approach that utilizes contrastive loss to improve classification and robustness against ASR errors.
Operational Barrier
Online teaching lacks standardization, leading to diverse instructional methods that challenge the system's effectiveness.
Proposed Solutions: Developing a flexible framework capable of adapting to various instructional styles through continuous learning and adaptation.
Data Quality Barrier
High transcription errors from ASR services can lead to inferior performance in detecting dialogic instructions.
Proposed Solutions: Using robust pre-trained language models to mitigate the impact of transcription errors during instruction detection.
Project Team
Yang Hao
Researcher
Hang Li
Researcher
Wenbiao Ding
Researcher
Zhongqin Wu
Researcher
Jiliang Tang
Researcher
Rose Luckin
Researcher
Zitao Liu
Researcher
Contact Information
For information about the paper, please contact the authors.
Authors: Yang Hao, Hang Li, Wenbiao Ding, Zhongqin Wu, Jiliang Tang, Rose Luckin, Zitao Liu
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