On the Similarity of Task Contexts

Developers work on parallel tasks and switch between them due to interruptions and dependencies. For each task, developers interact with artifacts that constitute the task context. The more dissimilar tasks are, the more time is needed for switches to restore the contexts and adjust the mindset. Organizing tasks by their similarity can increase the efficiency of task switches. Moreover, knowing similar tasks of other developers might also save coordination effort. This paper studies the concept of tasks similarity based on the similarity of task contexts. We define task context as set or vector of the artifacts used in the task and apply popular similarity measures to compare the sets or vectors corresponding to the tasks. Instead of only counting the artifacts in the intersection and union of two contexts, we scale the artifacts with their relevance to the task using the degree-of-interest model. In a field study, a simulation, and an experiment, we investigated the accuracy of the different context similarity models and compared them to the accuracy of text-based similarity using the task descriptions.