Pssst...or boo! assessing the predictability of notification delivery preferences

The focus of my dissertation research is on the examination of notification systems that harness different presentation formats for notification delivery, the preferences that individuals express for these various types of notifications, and how these preferences are affected by contextual information surrounding notification delivery. My research is unique from other work in the literature in two primary ways. First, while the majority of prior work addressing notification delivery, both in terms of format and timing, has focused on the effects of a notification on an individual's performance on a given task or set of tasks, my focus is the individual's perception of notifications, and particularly on that individual's preferences for different notification formats delivered within different contextual scenarios. An interest in this question is motivated by prior studies that have shown that annoyance with computer-human interactions is a primary reason behind user abandonment of interactive software systems [57, 88]. Second, my preliminary findings suggest that different people prefer different types of notifications in different contexts, which motivates a change of focus in the development of such systems toward customizing notifications not only to the features of an individual's context but also to the individual him- or herself. In this dissertation, in an attempt to model the ability of a machine attempting to customize its interactions to the user input it receives, individuality is measured at the level of revealed preferences; I do not attempt to understand the underlying individual differences that may also prove themselves to be factors behind the stated notification preferences examined here. I will describe four user studies I conducted to understand how preferences are exhibited in different contextual situations, detail the methodology followed and a set of guidelines for conducting similar studies (specifically in a rarely studied, naturalistic setting), and show how certain techniques from the area of statistics and user modeling can address the apparent need for personalized, context-sensitive notifications. An additional element of novelty in my work is that my final study was conducted in a purely naturalistic office environment, in which the notifications evaluated were precisely those notifications being delivered to study participants throughout their workday. The primary contribution of this dissertation is twofold: a detailed analysis of the methodology for the design, data collection, and analysis of a study of notification preferences in a naturalistic setting with a great deal of inherent complexity; and a set of results, based on the analysis of preference data acquired in various settings, about how an individual's contextual environment, and the content of a given notification, can affect that individual's preferences for notification delivery.