Web User Click Intention Prediction by Using Pupil Dilation Analysis

We propose a novel approach for predicting Web user click intention, using pupil dilation data generated by an eye-tracking device as input. Our goal is to determine if this variable is useful to differentiate choice and no-choice states, and if so, to generate a classification model for predicting choice understood as a click. For this, we performed an experiment with 25 healthy subjects in which gaze position and pupil size was recorded while users choose between several elements on a simulated Web site. Our results show that there is a statistical difference between pupil sizes of chosen elements compared with no chosen ones. Furthermore, we generated a click-intention prediction model, based on Artificial Neural Networks, which obtained an 82% accuracy. These results suggest that this variable could be used from a Web Intelligence point of view as a proxy of Web user behaviour, in order to generate an online recommender to improve Web site structure and content.