Describing User Actions in Adaptive Interfaces with Probabilistic Acceptors

Abstract

An important aspect of adaptive systems is the description of the user-system interaction, which can be used to derive new information about the user and to trigger adaptations, for instance by means of adaptation rules. In this paper, we present an approach that describes user actions by means of probabilistic deterministic finite- state automatons (PDFA), which are generated from an annotated corpus of user interactions. Based on a training set, different acceptors are created from recording data and can be employed by an adaptation framework to trigger adaptation rules. An evaluation of this approach with a prototype of an interactive TV system is presented.