Understanding of eye movements in reading web pages

Abstract

In this project, two websites LinkedIn and Facebook had been investigated in order to find out the strategies of usage of the information users. GMM model was used to find out the region of interests (ROIs). The model was then updated by Bayesian GMM to refine the number of clusters on each page and their location. Two experiments have been conducted on LinkedIn and Facebook. For LinkedIn, there are 40 subjects and they are not limited to any occupation and ages. For Facebook, 8 subjects are invited to do the experiment, they are all from Computer Science Department. In the LinkedIn experiment, subjects are asked to complete a by finding a recommendation link. Subjects are divided into groups, experienced/inexperienced, stressed/non-stressed. The fixations were first recorded in a list, the fixation data was then been classified by computing their velocity, fixations lower than 70 degree/s are considered as classified fixations. The consecutive fixations were then combined if they were close to each other. The analysis was done mainly based on the temporal graph and transition graph. A bunch of statistical tests have been used for the sake of finding the difference between different groups and tasks. Several surprising findings were found in this project. The first finding is that when people were asked to complete a task, people who were not under pressure tend to double check if they had done the task correctly. On the other hand, this will not happen on the group of people who are ask to complete the task as soon as possible. Another interesting finding was that experienced users would remember the position of webpage components such as advertisements. They would ignore the irrelevant information while performing the task. This behavior is considered as inattentional blindness. Finally, it is found out that users can remember their last position when they are switching from regions to regions. They would resume at the last position after they had finished the current one.