Multimodal Behavioral Disengagement Detection for Collaborative Game-Based Learning

Abstract

Collaborative game-based learning environments offer significant promise for creating effective and engaging group learning experiences. These environments enable small groups of students to work together toward a common goal by sharing information, asking questions, and constructing explanations. However, students periodically disengage from the learning process, which negatively affects their learning, and the impacts are more severe in collaborative learning environments as disengagement can propagate, affecting participation across the group. Here, we introduce a multimodal behavioral disengagement detection framework that uses facial expression analysis in conjunction with natural language analyses of group chat. We evaluate the framework with students interacting with a collaborative game-based learning environment for middle school science education. The multimodal behavioral disengagement detection framework integrating both facial expression and group chat modalities achieves higher levels of predictive accuracy than those of baseline unimodal models.