Streamlining the Cleaning and Analysis of Eye Tracking Data for Cross-Recurrence Analysis

The use of multimodal data to understand and support collaborative learning has grown in popularity recently as it allows researchers to study complex learning tasks from different facets. However, multimodal analysis is often computationally complex and often requires a strong set of technical skills and mathematical understanding to clean and process the data, creating barriers that restrict researchers without advanced programming skills from conducting these analyses. There are existing software packages that can help with this process, but they are often piecemeal requiring the user to understand how to put them together and what best practices may entail. This can create a barrier of entry for most researchers. Our toolkit addresses this challenge by providing researchers with an accessible, well-documented toolkit for synchronizing and analyzing multiple data streams from collaborative learning environments. In this paper, we present the first component of this toolkit: a comprehensive pipeline for preprocessing and analyzing collaborative eye tracking data, specifically focusing on cross-recurrence analysis. The pipeline automatically handles critical preprocessing steps including data synchronization, calibration filtering, and missing data management, then implements cross-recurrence analysis to measure gaze similarity between collaborating participants. Our toolkit will provide a step-by-step workflow in Jupyter Notebooks that requires no advanced programming skills. This tool will enable researchers to transform raw data into accurate predictions about collaborative behaviors, through a single interface. The toolkit generates real-time visualizations and handles the complex technical challenges that have previously limited access to these analytical techniques. While this paper focuses on cross-recurrence analysis as a foundational measure, future versions of the toolkit will incorporate additional collaborative measures such as spatial entropy, joint visual attention, and joint mental effort to provide a comprehensive suite of multimodal learning analytics tools.