eye-tracking analysis

Conventional eye-tracking analysis methods fail to characterize the temporal dynamics of visual attention during complex visual search tasks and cannot establish causal links between attentional behavior and task performance. Method: We propose a visualization analytics framework that jointly models oculomotor time-series patterns (fixations, saccades, smooth pursuits), object-click sequences, and multidimensional performance metrics (accuracy, efficiency). Leveraging time-series modeling and cognitive visualization techniques, the framework delivers interpretable representations of attention allocation and decision-making processes. Contribution/Results: Compared to isolated metric analysis, our framework significantly improves the fidelity of modeling attentional dynamics. It establishes, for the first time, a unified explanatory model integrating eye movements, interactive sequences, and task performance. Empirical evaluation in realistic, complex scenarios confirms its effectiveness in elucidating attentional mechanisms and optimizing human–computer interaction.

Traditional UX research relies heavily on subjective self-reports (e.g., surveys, interviews), limiting its ability to capture implicit user behaviors and cognitive states. To address this, we propose an objective, multimodal physiological assessment framework integrating eye-tracking (Tobii Pro Fusion) with biosignals—including electrodermal activity (EDA), heart rate variability (HRV), and pupil diameter—collected via Empatica E4. We introduce the first systematically validated, interpretable dual-dimensional metric for quantifying cognitive load and engagement, underpinned by an ethics-guided protocol for fusing heterogeneous physiological signals. Leveraging LabStreamingLayer for real-time synchronization and lightweight temporal modeling, our method achieves 91.3% accuracy (F1 = 0.89) in cognitive load classification and reduces engagement prediction error by 37% across 12 real-world interface evaluations. The resulting standardized operating procedure (SOP) for UX evaluation is publicly documented, reproducible, and has been adopted by three leading design teams.

This study investigates the identification of the Big Five personality traits from users’ eye-tracking behavior in an interactive museum environment. The work proposes a multimodal time-series model that, for the first time, treats periods of missing gaze data—intervals when eye tracking fails—as meaningful behavioral signals rather than noise. By integrating these missing segments with raw eye-movement data, the model captures the relationship between personality and visual search behavior while minimizing preprocessing to preserve naturalistic patterns. Ablation studies confirm that incorporating missing-data intervals significantly enhances discriminative power. Evaluated via five-fold cross-validation, the model achieves strong performance across all personality dimensions (Macro F1: 73.09%–77.69%), with the inclusion of missing-gaze information yielding a 10–15% improvement in both accuracy and F1 score.

Traditional assessments of visualization literacy rely solely on response accuracy, which fails to reveal the underlying cognitive processes and strategic differences users employ when interpreting charts. This study introduces, for the first time, a systematic framework that integrates eye-tracking methodology with standardized testing to construct process-oriented metrics—including component-level attention distribution, frequency of cross-region information integration, and dispersion of gaze trajectories. By moving beyond binary correctness judgments, this approach effectively captures variations in cognitive load and interpretation strategies that are invisible to conventional measures such as accuracy and response time. The proposed method thus enables a paradigm shift from outcome-based evaluation to fine-grained diagnosis of cognitive processes, offering a scalable and nuanced pathway for assessing visualization literacy.

This study addresses the challenge of modeling visual behavior and inferring intent during image browsing by users with low vision. Employing eye-tracking combined with retrospective think-aloud protocols, we conducted a comparative experiment across distinct types of visual impairment. We propose, for the first time, a five-category visual intent taxonomy specifically designed for low-vision users. Through qualitative coding and quantitative analysis, we uncover the synergistic interplay among visual acuity, image context, and oculomotor features—including fixation distribution and scanpath patterns. Results demonstrate that visual impairment type significantly modulates oculomotor correlates across intent categories; moreover, our taxonomy exhibits both interpretability and cross-group stability. This work provides foundational theoretical insights and empirical evidence to guide the development of intent-aware visual assistance technologies.

This study addresses the lack of systematic modeling of human–machine information interaction patterns in free-viewing eye-tracking data, which hinders the uncovering of deep associations between user behavior and stimulus characteristics. To this end, the paper proposes EnsembleGaze, an end-to-end unsupervised ensemble learning framework that introduces consensus subspace clustering and spectral biclustering to this domain for the first time. By leveraging statistical feature engineering of fixation distributions, EnsembleGaze performs consensus clustering on viewers, images, and their joint structures. The approach overcomes the limitations of traditional unidimensional analyses, demonstrating on public datasets that image groupings exhibit high consistency—reflecting stable ambient–focal viewing modes—whereas viewer groupings are context-dependent and can only be effectively recovered through joint modeling strategies such as biclustering.

This study addresses a critical gap in existing eye-tracking research on gaming, which has predominantly focused on fixation locations and durations while overlooking transition patterns between functional regions and their relationship to gameplay performance. Integrating three classes of oculomotor metrics—fixation distribution, duration, and transitions—the authors employ a within-subjects experimental design to analyze players’ attention allocation and switching behaviors across six functional areas of interest (AOIs) in a card-battle interface, linking these patterns to win–loss outcomes. The findings reveal, for the first time in strategic games, that superior performance is associated with more balanced attentional distribution, higher-frequency action-oriented transitions, and greater diversity in AOI combinations, collectively indicating enhanced capacity for information coordination.

This study investigates how ensemble musicians achieve coordination through visual attention in natural rehearsal settings. Using Pupil Labs Neon mobile eye trackers, the researchers recorded a quartet during rehearsals and applied YOLOv8 for scene annotation, followed by multimodal analysis integrating gaze matrices, transition matrices, temporal scarf plots, and interview data. The findings reveal, for the first time, a “hub-and-spoke” attention topology during rehearsal: the lead performer serves as the central gaze target, receiving 97% of interpersonal fixations from a novice guitarist. Repetitive practice reduces gaze transitions by an average of 65% (up to 82%), indicating dynamic stabilization of attentional focus, whereas pedagogical interruptions induce attentional fragmentation, while uninterrupted performance fosters integration.

This study addresses the degradation of operational effectiveness in high-risk C5ISR command environments caused by suboptimal attention allocation. Conducted within a high-fidelity military command simulation, it pioneers the application of recurrence quantification analysis (RQA) to model dynamic attention patterns, integrating eye-tracking data with multimodal adaptive decision support tools to investigate the relationship between gaze dynamics and task performance. Results demonstrate that multimodal guidance significantly outperforms purely visual guidance. Specifically, mean diagonal line length exhibits a negative correlation with performance, while entropy shows a positive correlation. Moreover, recurrence rate and determinism follow an inverted U-shaped relationship, aligning with the Yerkes–Dodson law. These findings uncover a nonlinear mechanism linking gaze patterns to task performance, offering theoretical and methodological foundations for intelligent command decision-aid systems.

This study addresses a gap in eye-tracking research by examining visual attention during art viewing, an area largely overlooked in favor of social scenes. It proposes a novel analytical framework integrating spatial (fixation density maps) and temporal (scanpaths) dimensions to systematically compare visual exploration patterns among autistic individuals, artists, and neurotypical controls during free viewing of 30 paintings. Using dispersion-threshold identification to extract fixations and evaluating performance with six saliency metrics—including AUC-Judd and NSS—alongside temporal alignment methods such as MultiMatch and Dynamic Time Warping, the study reveals high consistency between artists and neurotypicals, whereas autistic participants exhibit broader spatial exploration, shorter fixation durations, and highly idiosyncratic scanpaths. These findings underscore a distinct aesthetic attention mechanism in autism and advocate for population-specific modeling of visual aesthetic attention.