[Iqbal 06]

Iqbal ST, Bailey BP. Leveraging characteristics of task structure to predict the cost of interruption. Proceedings of the SIGCHI conference on Human Factors in computing systems - CHI '06. 2006:741. Available at: http://portal.acm.org/citation.cfm?doid=1124772.1124882.

• goal: create parsimonious model that uses characteristics of task structure to predict COI, a model that can be expediently applied to many goal-directed tasks

• systems could predict a more accurate COI if they also considered characteristics related to the structure of a task - subtasks, boundaries within task decomposition
• prior work showed that interrupting at subtask boundaries results in much lower COI than non-boundary moments
• users performed representative primary tasks and were interrupted at various boundaries with peripheral tasks; resumption lag, time to resume primary tasks after an interruption, was used to provide ground truth for COI
• benefit of COI model is that it can be expediently applied to approximate COI at subtask boundaries in many goal-directed tasks, without a physiological measure of workload

• empirical COI: interrupting at random moments can cause users to take up to 30% longer to resume the tasks, commit up to twice the errors, experience up to twice the negative affect than when interrupted at boundaries
• interrupting at earlier phases in task structure had lower cost;
• current work focus on subtask boundaries because systems can detect them, further investigates which characteristics of a task's structure best predict the COI at subtask boundaries;
• predicting COI: build a probabilistic model w/ input cues including desktop activity, visual and acoustical analysis, scheduled activities of the user

• experiment 1: uses a within-subjects design
• initial models based on authors' own understanding of tasks' execution; each model achieved more than 90% accuracy with no obvious patterns in the errors
• peripheral tasks took ~20s to perform and must be responded to as quickly as possible, while returning to the primary task as quickly as possible
• moments for interruption chosen from sample of ten representative subtask boundaries from corresponding GOMS model
• measurements: resumption lag = COI
• candidate factors for predicting COI: level, presence (not saliency) of visual resumption cue, percent of task complete at boundary, percent of parent subtask complete at boundary, difficulty of preceding subtask, difficulty of next subtask, carry over at boundaries;
• determining most predictive factors of resumption lag: at least one of the factors has a non-zero coefficient: level, carry-over, difficulty of next subtask; COI depends more on characteristics that reflect current and prospective allocation of mental resources than on those that reflect temporal position or cue availability in the task
• decided to cluster resumption lag into classes and determine cost for these interrupt-ability classes; total # of correct predictions 63.2%, much higher than chance (33%)

• experiment 2: eval. of the COI model; assignment of subtask boundaries to 3 COI clusters - correctly predicted 53% of COI values, much better than chance (33%); COI model can be reasonably generalised to other goal-directed tasks
• validates that a system can and should use our model to differentiate among subtask boundaries, enabling more effective decisions about when to interrupt

• discussion: model can differentiate subtask boundaries based on workload;
• limitations: model considers only one component of task structure - subtask boundaries, because systems can detect them
• familiarity with a task seems to have little effect on how users perceive its hierarchical structure; a possible solution is to extend COi model to include skill level as a predictor
• COI model best suited for tasks that are high freq., routine, safety critical

• future work: investigate automated methods for building task models and predicting COI
• extend COi model to include non-boundary points
• implement our COI model within an existing interruption reasoning system

• Adamczyk PD, Bailey BP. If not now, when?: the effects of interruption at different moments within task execution. of the SIGCHI conference on Human. 2004;6(1):271-278. Available at: http://portal.acm.org/citation.cfm?id=985692.985727.