Unveiling Cognitive Strategies: Clustering Eye-Tracking Measures in Computerized Matrix-Reasoning Tasks

Cognitive strategies are interesting in their malleability and essential as fundamental drivers of reasoning success. Based on the eye-tracker enhanced computerized matrix-reasoning tasks, while data mining algorithms have been employed to unveil various visual search patterns in reasoning tasks, their success in identifying cognitive strategies has been partial, possibly due to not analyzing suitable data that can reflect together cognitive strategies. This study utilizes two clustering algorithms, K-means and K-prototypes, to categorize groups based on distinct cognitive strategies—construct matching and response elimination—using validated eye-tracking measures: proportional time on matrix area, rate of toggling, and rate of latency to first toggle. Key findings include (1) optimal categorization into two clusters for most items, aligning with established cognitive strategies; (2) the predominant influence of eye-tracking measures in strategy inference, with limited impact from additional response accuracy data; (3) superior performance of participants employing construct matching over response elimination in the majority of items; and (4) a tendency for participants to prefer construct matching on easier items. Overall, the study showcases the viability of data mining algorithms in identifying cognitive strategies through the analysis of pertinent eye-tracking measures.