Soazig Casteau¹ (), Alice Nevard¹, Ulrik Beierholm¹, Holger Wiese¹; ¹Durham University

Human social interaction is highly dependent on effective visual attention, yet current eye-tracking research largely relies on static stimuli or single-session analyses. This limits our understanding of whether individual social viewing patterns are consistent, persistent strategies that stay over time, or if they are dependent on short-term context and individual adaptation. To address this, we tested whether individual viewing strategies remain invariable across multiple, naturalistic social interactions. We (1) used Hidden Markov Models (HMMs) to identify probabilistic social viewing strategies across multiple conversational sessions; and (2) determined if the adoption of a specific HMM-defined strategy is consistent over time and predictive of behavioural outcomes. Forty participants engaged in a free-form social conversation with an unfamiliar partner (to-be-learned face) during four 10-minute-long sessions spaced over a one-week period. Eye movements were recorded using mobile eye-tracking, and face recognition accuracy of the partner was measured after each session to assess learning. Consistent with previous findings, the HMM analysis revealed two probabilistic viewing strategies: Analytic and Holistic. Crucially, our results showed that each participant’s dominant viewing strategy remained consistent across sessions, demonstrating stable individual differences independent of session progression. While participants improved their face recognition accuracy over time, overall rate of learning was similar between both strategy groups. Participants predominantly adopting the Holistic strategy showed higher initial face recognition accuracy, but this early difference converged quickly across sessions, and strategy type was ultimately not a reliable predictor of long-term recognition performance. Taken together, our findings establish that an individual's social viewing strategy is a highly persistent characteristic independent of time and accumulating experience. Although an initial performance difference was observed early in the task, this consistent strategy wasn't related to long-term recognition performance and the overall rate of learning, suggesting a limited functional role in the context of this specific naturalistic social task.

Acknowledgements: This research has been supported by an ESRC grant (ES/X014657/1)