This is a project to develop a new approach to interruptibility and related issues in instant messaging (IM) and mobile communications. Computer mediated communication is hampered by impoverished social cues related to when it is appropriate for an information system to interrupt a person or carry out some other action that might be appropriate at some times and not others. This problem is more pronounced when computing becomes more pervasive and as people are increasingly always online and always reachable. In the case of IM, the old method of alerting users to one's status no longer functions well. Although computers have access to many embedded sensors, automated assistance for alleviating inappropriate IM interruptions remains elusive. In a world in which users are ""available"" for some purposes, but not for others, one indication of interruptibility no longer fits the needs of users.
This research will explore how sensor data coupled with machine learning techniques and mass-collaboration could be leveraged to support users' social decisions. By learning users' preferred description of their place - in light of their position, activity, time, etc. - an intelligent user interface could present communication partners with information about a user's current status, ideally their ""context."" This shifts the burden of determining interruptibility from the realm of the computer to the social realm of the users, where it belongs. This approach treats context as an evolving communication of environmental data. Central to this approach is a way in which users are motivated to submit training data that translates sensor data to semantic labels.
The work will be carried out in three phases. Phase I: Development, deployment and testing of a context-aware IM client, Nomatic*Gaim. This client will be used as a data collection mechanism for mappings of sensor data to semantic labels that will train machine learning algorithms. Under the auspices of a controlled user study, this software will be deployed, and researchers will observe the language that humans use to disclose context. Phase II: Organic ontologies will be constructed from correlations in the data set using statistical tests. These ontologies will form a language structure on which the vocabulary of Phase I can be placed. The models from Phase I coupled with the structure learned in Phase II will be incorporated into a user interface that assists users with setting their context status and that gracefully degrades as confidence in predictions gets lower. Evaluations of this phase will be conducted using user-interface evaluation techniques that will quantitatively and qualitatively judge the effectiveness of automatic labeling. Phase III: The newly developed IM techniques will be incorporated into a mobile phone version of Nomatic*Gaim that will then be evaluated in a parallel study to Phase I. This will enable researchers to understand differences between laptop and cell-phone mobility.
This research will push the boundaries of effective social communication in an ""always online"" culture. Effective semantic context labeling promises to have a broader impact on computing by becoming a resource which prompts software to act in new ways. For example, when a user is in a library the volume of their device can be lowered; when a user is on a boat, tide tables and maritime weather reports can be automatically obtained and displayed more prominently. The software and research artifacts will be incorporated into existing undergraduate research plans and will help to seed the development of computer science curriculum organized around the idea of computing with location.
