Reasoning about ideal interruptible moments: A soft computing implementation of an interruption classifier in free-form task environments

Current trends in society and technology make the concept of interruption a central human computer interaction problem. In this work, a novel soft computing implementation for an Interruption Classifier was designed, developed and evaluated that draws from a user model and real-time observations of the user's actions as s/he works on computer-based tasks to determine ideal times to interact with the user. This research is timely as the number of interruptions people experience daily has grown considerably over the last decade. Thus, systems are needed to manage interruptions by reasoning about ideal timings of interactions. This research shows: (1) the classifier incorporates a user model in its’ reasoning process. Most of the research in this area has focused on task-based contextual information when designing systems that reason about interruptions; (2) the classifier performed at 96% accuracy in experimental test scenarios and significantly outperformed other comparable systems; (3) the classifier is implemented using an advanced machine learning technology—an Adaptive Neural-Fuzzy Inference System—this is unique since all other systems use Bayesian Networks or other machine learning tools; (4) the classifier does not require any direct user involvement—in other systems, users must provide interruption annotations while reviewing video sessions so the system can learn; and (5) a promising direction for reasoning about interruptions for free-form tasks–this is largely an unsolved problem.