A comparison between decision trees and Markov models to support proactive interfaces

Abstract

During the past few years, personal portable computer systems such as PDAs or laptops are being used in different contexts such as in meetings, at the office, or at home. In the current era of multimodal interaction, each context may require other interaction strategies or system settings to allow the end-users to reach their envisioned goals. For instance, in a meeting room a user may want to use the projection equipment and disable the audio output for a presentation, while audio input and output may be important while in a teleconference. In present computer systems most changes have to be made manually and require explicit interaction with the system. The number of different devices used in such environments makes that this configuration step results in a high cognitive load and causes interrupts of the tasks being executed by the end-user. In this paper we present how proactive user interfaces may predict the next interface changes invoked by context switches or user actions. In particular, we will focus on two machine learning algorithms, decision trees and Markov models, that may support this proactive behaviour for multimodal user interfaces. Based on some simple but relevant scenarios, we compare the outcome of both implementations in order to decide which algorithm is most applicable in this context.