Cascaded generic XCS to learn about reminding preferences

Nadine Richard, Samuel Tardieu, Seiji Yamada

Abstract: We are developing an adaptive reminding system, which learns when and how to present notifications. In this paper, we focus on our XCS-based model, composed of two cascaded sets of classifiers: the first one learns a categorization of calendar data, while the second selects the appropriate forms of combinable reminders depending on the user and device contexts. After describing the characteristics of the input data, we present the extensions we propose to provide a generic XCS architecture, which seems suitable for processing those specific inputs. Finally, we describe our user feedback mechanism, and the according reward system.