Grouping tasks and data display items via the non-negative matrix factorization

Analyzing work functions and the IO variables they need is an important component of designing and evaluating complex systems. We develop a biclustering method for jointly grouping work functions and IO variables. Given a binary matrix indicating which IO variables are required for which work functions, we develop a computational method for finding dense groups of related tasks and information. We relax the binary grouping problem to a continuous one, and then use the nonnegative matrix factorization, followed by dichotomization, to search for useful groupings. The result is a more automated approach to finding groupings for interaction design. Our motivating data set comes from NASA, where the tasks are those that a pilot must perform and the sources are displays required for those tasks.