Skip to main content Skip to navigation

Rethinking the Potential of Multimodality in Collaborative Problem Solving Diagnosis with Large Language Models

Project Overview

The document explores the transformative role of generative AI, particularly large language models (LLMs) and multimodal data, in enhancing the assessment of Collaborative Problem Solving (CPS) skills within education. It highlights research demonstrating that transformer-based models, especially those leveraging both text and audio inputs, significantly outperform traditional assessment methods in identifying indicators of CPS. Despite these advancements, the findings indicate that the efficacy of multimodal approaches can vary based on the specific characteristics of the CPS indicators being evaluated. Furthermore, the paper underscores the critical role of human-AI collaboration in educational settings, suggesting that combining AI capabilities with human insights can lead to more effective and nuanced assessments of student skills. Overall, the document presents a promising outlook on the integration of generative AI in education, advocating for a balanced approach that harnesses the strengths of AI while emphasizing the irreplaceable value of human expertise.

Key Applications

Multimodal classification model for diagnosing collaborative problem solving (CPS) subskills

Context: Secondary education, specifically targeting students aged 14-15 in mathematics classes

Implementation: Utilized transformer-based models with multimodal data (text embeddings from verbal data and acoustic embeddings from audio data) to automate the diagnosis of CPS indicators

Outcomes: Improved detection of CPS classes with transformer-based models, especially for social-cognitive CPS classes

Challenges: Integration of multimodal data did not universally improve model performance; complexity of CPS indicators and data composition affected outcomes

Implementation Barriers

Technical

Challenges in accurately detecting CPS indicators in noisy classroom environments and the reliance on manual coding of indicators.

Proposed Solutions: Automated detection using analytics and machine learning techniques, exploring ensemble learning methods for model improvements.

Data-related

Limited empirical work on the potential of LLMs and multimodal data for CPS diagnosis, particularly in real-world scenarios.

Proposed Solutions: Encouraging more rigorous empirical studies and cross-validation with diverse datasets to enhance generalizability.

Project Team

K. Wong

Researcher

B. Wu

Researcher

S. Bulathwela

Researcher

M. Cukurova

Researcher

Contact Information

For information about the paper, please contact the authors.

Authors: K. Wong, B. Wu, S. Bulathwela, M. Cukurova

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