Characterizing joint attention dynamics during collaborative problem-solving in an immersive astronomy simulation

The complex and dynamic nature of collaboration makes it challenging to find indicators of productive learning and quality collaboration. This is particularly the case when learners were situated in an exploratory simulation that typically generates large-scale unstructured logs. Joint attention (JA) plays an important role in understanding how students coordinate their attention with others in collaborative problem-solving tasks. While previous literature supports the correlation between higher-level JA and learning performance, it does not tell the whole story as JA may hide different collaboration behaviors including imbalance participation known as the free-rider effect. In addition, JA was studied mostly through qualitative observations and eye trackers. This study, therefore, utilizes logs generated as students interact with an immersive astronomy simulation using augmented reality head-sets (Microsoft HoloLens 2) and tablets. We aim to explore JA behaviors to investigate collaborative problem-solving patterns across different learning performance groups. We developed a metric to capture different JA states, with a specific focus on one key behavior: consistent screen overlapping across devices as a proxy for JA. We then investigated sequences of these states to explore dynamic collaboration trajectories across groups by examining two transition metrics: Markov chain model and L*. Our results show a higher level of JA is associated with quality collaboration and contributes to learning gains, which supports the findings from the previous studies. One interesting finding is a longer consistent screen overlapping (i.e., long-shared view state) tendency in the early stage for high-learning-gain groups. Together, our findings indicate the potential of JA metric to predict overall collaboration quality and serve as an early warning for instructors to provide in-time guidance