A Bayesian approach to predicting website revisitation on mobile phones

Abstract

Mobile web browsing is highly recurrent, in that a large proportion of user's page requests are to a small set of websites. Despite this, most mobile browsers do not provide an efficient means for revisiting sites. Although significant research exists on prediction in the personal computer realm, little work has been done in the mobile realm where physical constraints of the device and mobile browsing behaviors are vastly different. The current research proposes a Bayesian model approach, based on a cognitive model of memory retrieval that integrates multiple cues in order to predict the next site a user will visit. These cues include frequency of site visitation, the recency of site visitation, and the context in which specific sites are accessed. The model is assessed using previously collected web logs from 24 iPhone users over the course of one year. Our model outperforms simpler models based on frequency or recency, which are sometimes implemented in desktop browsers. Potential applications of the model are discussed with the objective of increasing browsing efficiency on mobile devices. HighlightsPredicts website revisitation on mobile phones.Accounts for nearly 50% of site visits.Incorporates frequency, recency, and context (current site) to make predictions.Comparison of models to show contribution of each predictive cue.