Predicting Web User Click Intention Using Pupil Dilation and Electroencephalogram Analysis

In this work, a new approach for analysing the Web user behavior is introduced, consisting of a physiological-based click intention assessment, based on pupil dilation and electroencephalogram (EEG) responses evaluation. For this, an empirical study was conducted, where the mentioned responses of 21 subjects were recorded while performing diverse information foraging tasks from five real web sites. We found a statistical difference between click and not-click pupil dilation curves, more precisely, fixations corresponding to clicks had greater pupil size than fixations without clicks. In addition, seven classification models were applied, using 15 out 789 pupil dilation and EEG features obtained from a Random Lasso feature selection process. Results showed good performance for Accuracy (71,09% using Logistic Regression), whereas for Precision, Recall and F-Measure remained low, which indicates the behavior we were studying was not well classified. Despite the quality of these results, it is possible to mention that the reviewed responses could be used from a Web Intelligence perspective as a proxy of Web user behavior, for example, to generate an online recommender to improve websites structure or content. However, we concluded that better quality instruments are necessary to achieve higher results.