Bayesian and behavior networks for context-adaptive user interface in a ubiquitous home environment

Modern home theater systems require users to control various devices simultaneously including a TV, audio equipment, DVD and video players, and a receiver. To perform the requested user functions in this situation, the user is required to know the functions and positions of the buttons on several remote controls. Users will become more confused if a ubiquitous home environment, which contains many mobile and stationary control devices, is realized. Therefore, the user interface should be adaptable for requested user functions and to fit a specific control device. This paper presents a context-adaptive user interface for the control of devices in ubiquitous home environment. First, we modeled the ubiquitous home environment in order to implement the context-adaptive user interface. We used a Bayesian network to predict the necessary devices in each situation and used a behavior network to select the functions that constitute an adaptive user interface in several conditions. The selected functions were used to generate an adaptive interface for each controller using a presentation template. In this paper, we implemented a ubiquitous home environment and generated a controller usage log for this environment. We confirmed that the Bayesian network effectively predicted the user requirements by evaluating the inferred results of the necessary devices based on several scenarios. Finally, we compared the adaptive user interface with the fixed user interface by surveying fourteen subjects. We confirmed that the generated adaptive user interface was more comfortable for use with typical tasks than was the fixed user interface.

► This paper presents a work on using Bayesian and behavior networks to generate an adaptive user interface for smart home environments.
► The needs come from the complexity of device control in a smart home environment.
► This paper proposes to use context information to infer more relevant controls.
► The approach is a mix of induction-based (machine learning) and deduction-based (rule inference).
► We compared the adaptive user interface with the fixed user interface by surveying fourteen subjects.