On the Use of Predictive Deep Learning Approaches in the Frequency and Uniqueness-Based Representation of Sequential Browsing Events

Abstract

To gain a more comprehensive understanding of how users navigate, this study investigates the potential of predictive deep learning techniques. To properly describe user's browsing events such page visits, this paper first provides an overview of two embedding methodologies based on natural language processing and graphical representations. Secondly, this study presents two aggregation methods, one based on frequency and the other on uniqueness, that can be used to depict the possible sequence of events that may occur during a user's surfing session. With an emphasis on online banking, this research summarizes data representation approaches to model customers' digital footprints and learn about their banking product purchase intents. When we analyze a customer's digital footprint in the digital banking domain during the early stages of the application's lifecycle, we find that there are insufficient digital user footprints to model and comprehend users' behavior. Consequently, new approaches are needed to increase the quantity of session recordings kept by customers' digital footprints. We consider the possibility of increasing the quantity of digital footprints of bank customers by employing predictive deep learning techniques. To enhance the quantity of user navigational behaviors, we present a predictive technique based on long short-term memories. According to the findings, the newly generated customer sequences facilitate testing the suggested method, and accurately represent the initial client session data.