Improvements to keyboard optimization with integer programming

Keyboard optimization is concerned with the design of keyboards for different terminals, languages, user groups, and tasks. Previous work in HCI has used random search based methods, such as simulated annealing. These "black box" approaches are convenient, because good solutions are found quickly and no assumption must be made about the objective function. This paper contributes by developing integer programming (IP) as a complementary approach. To this end, we present IP formulations for the letter assignment problem and solve them by branch-and-bound. Although computationally expensive, we show that IP offers two strong benefits. First, its structured non-random search approach improves the out- comes. Second, it guarantees bounds, which increases the designer's confidence over the quality of results. We report improvements to three keyboard optimization cases.