<p>Sports videos serve as critical resources for athletes and coaches to observe and replicate expert performance, yet current viewing practices lack quantitative tools to assess visual attention alignment with key body movements. To address this, we propose GazeSport, a visual analytics framework that integrates high-precision eye tracking with human pose estimation. GazeSport synchronizes gaze fixations and athlete key points to generate dynamic areas of interest (AOIs) using a region growing algorithm based on pose landmarks, adapting to motion in real time. It visualizes gaze behavior through spatiotemporal representations, allowing users to examine attention on relevant body parts and detect distractions toward nonessential elements such as referees and equipment. Viewing efficiency is evaluated using gaze entropy and Bayesian uncertainty. Designed for sports involving intensive body movement and suitable for individual or multi-athlete scenarios, GazeSport has been validated through a user study involving 18 volunteers with diverse professional backgrounds viewing the sports video from the SportsSloMo Dataset. Results reveal clear differences in attention efficiency among participant groups, demonstrating the effectiveness and generalizability of the proposed framework.</p> Graphical Abstract <p></p>