Task abstractions and taxonomic structures for tasks are useful for designers of interactive data analysis approaches, serving as design targets and evaluation criteria alike. For individual data types, dataset-specific taxonomic structures capture unique data characteristics, while being generalizable across application domains. The creation of dataset-centric but domain-agnostic taxonomic structures is difficult, especially if best practices for a focused data type are still missing, observing experts is not feasible, and means for reflection and generalization are scarce. We discovered this need for methodological support when working with time-stamped event sequences, a datatype that has not yet been fully systematically studied in visualization research. To address this shortcoming, we present a methodology that enables researchers to abstract tasks and build dataset-centric taxonomic structures in five phases (data collection, coding, task categorization, task synthesis, and action-target(criterion) crosscut). We validate the methodology by applying it to time-stamped event sequences and present a task typology that uses triples as a novel language of description for tasks: (1) action, (2) data target, and (3) data criterion. We further evaluate the descriptive power of the typology with a real-world case on cybersecurity.
View on arXiv