Conversational Context for Mobile Notification Management

This thesis explores if and how identifying the character of face-to-face conversations can help manage proactive actions, especially notifications, on smartphones so that they become less disruptive. We motivate our research with the initial hypothesis that different types of conversation entail a different receptiveness to interruptions. This hypothesis is backed up by surveys and eventually proved in a large-scale user study. We show that the social dimensions depth/importance and formality/goal orientation of a conversation are strong indicators of receptiveness. We also provide evidence that, contrary to initial belief, valence is no indicator by itself. Furthermore, we find that there are types of conversation, small talk for example, in which individuals are even more receptive to notifications than in situations without any verbal social interaction at all. This refutes the assumption currently dominating the literature that the occurrence of a conversation is a strong predictor of unavailability. The study also reveals that 44% of the time a message is received, people are engaged in a conversation, attributing further importance to the consideration of conversational characteristics in notification management. We investigate how and to what degree the character of a conversation can be inferred by technical means – putting strong emphasis on the challenges of mobile scenarios. For that, we study how state-of-the-art methods in speech-based machine learning cope with the effects of noise and signal alterations through the manner the phone is carried by the user. We demonstrate a system that keeps track of conversations in which the user is engaged and that analyzes speech in terms of embedded affective and social cues. Eventually, we find that information of either kind, derived from audio, improves the accuracy of personal notification preference models for the average individual by more than 20 percent (relative to the baseline set of common context attributes). Funding for this research project was provided by the German Federal Ministry of Education and Research under the grant ID 01IS12057. The sole responsibility for the report’s contents lies with the author.