On detecting user’s attention from physical activities for attention-based interfaces

Abstract

Modeling user’s attention state can immensely help in the development of attentive device (s) that can proactively present user with correct information when and where needed. Detecting how user attends to devices and interfaces is a crucial problem in user interface design; especially for ubiquitous systems. To help inform the design of such systems, it is important to know how a user pays attention to an interface. Gaze was used in several researches to determine user’s attention, but we argue that gaze is not always practical, especially if the communication is non-visual. In this paper, a method that detects and infers user’s attention from activity information obtained from body-worn sensors was used. Three different experiments were conducted to investigate the effect of varying audio signals to user’s attention, and a classification system based on Gaussian naïve bayes method was implemented in MATLAB for detecting change in user activity pattern. The correlation between such changes and the occurrence of an audio signal was shown. The System was evaluated, and an overall accuracy of 80.04% from experiments 1 and 2, and 79.36% for experiment3 was obtained.  Based on the result and the accuracy of the classifier in detecting user’s points of attention, it was concluded that body-worn sensors could be used to detect attention to audio interface from user’s activity; hence physical activity can be used for Attention-based interfaces.