LLM-Generated Knowledge Graphs for Analyzing the Association between Student Participation and Learning in Online Math Discussions

• In Event: AERA Poster Session Wednesday 1:45pm
  In Poster Session: Contemporary Issues in Online Instruction and Student Learning

Abstract

This study introduces a novel method for analyzing student learning by combining social network analysis with large language model (LLM)-generated knowledge graphs (KGs) in a human-in-the-loop framework. Using over two million online math discussion records, students were classified into core, peripheral, or extra-peripheral roles based on their interaction patterns. Their knowledge structures were assessed through the Selection-Organization-Integration (SOI) model applied to the LLM-generated KGs. Results showed that core students demonstrated deeper knowledge SOI and benefited more from tutor feedback than x-peripheral peers. X-peripheral students, with limited interaction, faced challenges in structuring knowledge. These findings demonstrate the utility of LLM-generated KGs for assessing learning at scale and highlight the need for role-sensitive supports in online learning environments.

Authors

• Chenglu Li, University of Utah

• Bailing Lyu, University of Utah

• Gokhan Gulfidan, University of Utah

• Linlin Wu, University of Utah

• Rui Guo, University of Florida