Risk decision-making behavior in heritage tourism destinations based on eye-tracking experiments in Gulangyu, China

With the increasing frequency and severity of both natural and man-made disasters, risk decision-making behavior (RDMB) in heritage tourism destinations (HTDs) has gained growing attention. Traditional studies lack a comprehensive quantification from a “human-centered” perspective and ignore individual perceptual differences in earthquake scenarios, particularly in HTDs with complex architectural layouts and diverse road systems. Taking Gulangyu as a case study, this paper integrates behavioral decision theory (BDT) with the Stimulus-Organism-Response (SOR) theory to construct a model for analyzing RDMB. The Santa Barbara Sense of Direction Scale (SBSOD) was employed to assess participants’ directional ability, and video stimuli were used to evoke participants’ risk perceptions. A Fully Convolutional Network (FCN) was applied to semantically quantify street view images (SVIs), while eye-tracking technology was used to monitor participants’ eye movements, physiological responses, and behaviors. Additionally, a multiple linear regression model was utilized to analyze the influence of street view elements (SVEs) on RDMB. The results show that (1) in terms of psychological cognition, individuals with stronger sense of direction exhibited fewer hesitation behaviors and more stable heart rates. (2) regarding physical space, the elements affecting decision-making behavior in street views were ranked as follows: buildings, roads, sky, vegetation, other elements, and signage. (3) in terms of physiological response, participants primarily used buildings and signage to identify their location, followed by vegetation and the sky for environmental observation. Theoretically, this study extends the application of SOR theory and BDT and proposes an analytical path involving psychological cognitive stimulation, physical space quantification, and physiological response (P-P-P). Methodologically, it integrates eye-tracking technology with deep learning (DL) algorithms. Practically, this study provides a basis for improving disaster resilience, enhancing tourist safety perception, and increasing the efficiency of emergency response in HTDs.