Multimodal Data for Learning Regulation with AI in Collaborative Learning (LEAD-UMLA-AI-PRCL)

This dataset comprises multimodal data collected from 52 pre-service teachers who participated in a 60-minute collaborative inquiry learning task on the topic of forms and conservation of energy. The participants were divided into 14 groups, each consisting of 3-4 members. This dataset is part of a joint initiative between two Research Council of Finland-funded projects: (1) "Utilising Multimodal Learning Analytics with Artificial Intelligence (AI) to Predict Regulation in Collaborative Learning (UMLA-AI-PRCL)" (P.I. Andy (Khánh) Nguyen), and (2) "Learning Regulation with AI – Promoting Adaptive K-12 Learners (LEAD)" (P.I. Sanna Järvelä).

The data collection involved various modalities, including 2D camera video, individual and group audio recordings, close-up webcam videos capturing facial expressions, Kinect depth-view videos, screen recordings, log data, self-reports with the Partner Model Questionnaire (Doyle, 2022) and post-interviews, and physiological measurements such as heart rate and electrodermal activity (EDA).

The collaborative learning activity was structured using a learning script that included an introductory video (5 minutes) on the topic, a simulation exercise with guiding questions and an answer sheet (University of Colorado Boulder, 2024; 20 minutes), and an experimental task focused on the forms and conservation of energy, specifically observing how different balls bounced on the floor (25 minutes). After the task, participants engaged in a reflective discussion on how such activities could be integrated into their future teaching practices (10 minutes). All activities and instructions were implemented within the GoLab learning environment (Golabz, 2020).

During the task, students interacted with MAI (Metacognitive Artificial Intelligence) in two ways. They could ask task-related questions via text messages using a chatbot interface powered by OpenAI’s GPT-4, which was designed to avoid providing direct answers and instead encouraged reflection and the exploration of alternative problem-solving strategies. Additionally, MAI monitored group discussions and provided Self-Regulated Learning (SRL) prompts via a speaker to support group collaboration. To manage the challenges of detecting speech accurately in authentic collaborative learning contexts, the Wizard of Oz (WOz) paradigm was used to simulate this interaction modality.