#947: Assessing Student Collaboration with Multimodal Data and Large Language Models

We introduce a multimodal evaluation approach that leverages large language models to assess group collaboration quality in computer-supported collaborative learning environments. Using synchronized log data, dialogue transcripts, and human-coded annotations from undergraduate dyads working in the Graspable Math platform, we develop a theoretically grounded rubric capturing cognitive, social, and regulatory dimensions of collaboration, and examine the extent to which LLM-generated evaluations align with expert human judgments.