Session-based path prediction by combining local and global content preferences

Abstract

Session-based future page prediction is important for online web experiences to understand user behavior, pre-fetching future content, and for creating future experiences for users. While webpages visited by the user in the current session capture the users’ local preferences, in this work, we show how the global content preferences at the given instant can assist in this task. We present DRS-LaG, a Deep Reinforcement Learning System, based on Local and Global preferences. We capture these global content preferences by tracking a key analytics KPI, the number of views. The problem is formulated using an agent which predicts the next page to be visited by the user, based on the historic webpage content and analytics. In an offline setting, we show how the model can be used for predicting the next webpage that the user visits. The online evaluation shows how this framework can be deployed on a website for dynamic adaptation of web experiences, based on both local and global preferences.