Directional scan path characterization of eye tracking sequences: A multi-scale approach

Abstract

Historically, eye tracking systems have been a very useful tool for finding salient regions in interfaces that naturally attract the visual attention of users. Scan paths are created as the eye moves from one salient region to another. Research has shown that a relationship exists between scan path direction and cognitive load when navigating a user interface. The analysis of scan paths during interface navigation can therefore indicate specific user tendences that could give insight into the cognitive load required to navigate an interface. Thus far, the analysis of the scan paths has largely focused on measuring scan path similarity or analyzing scan paths and their relationship to a predetermined AOI sequence. The focus of this paper is the characterization of eye tracking results based on the analysis of scan path sequences. The proposed approach summarizes scan paths at a macroscale associated with the overall navigation of an interface and a microscale paths associated with saccades that occur within fixations. The approach uses nearest neighbor search and density-based clustering to differentiate between the macroscale and microscale. A directional sequence encoding technique based on the saccade angle is then applied at each level. Finally, frequent sequence mining is used to find frequent scan paths at both scales. Empirical results show that the resulting set of frequent sequences offer an interesting summary of summarize the users scan path for an eye tracking session and can be used to indicate information seeking tendencies of users and characterizes visual search patterns.