Automatically generating personalized user interfaces with Supple
Download 5.78 Mb. View original pdf
|
| 5. Formulating the cost function The style and quality of user interfaces generated by Supple is determined by the cost function, which provides a quantitative metric of user interface quality. In this section, we develop two different cost functions. The first is factored in a manner that enables fast computation of an admissible heuristic. This cost function is also parametrized in such away that different choices of parameters can result indifferent styles of user interfaces generated. Subsequent work [28] demonstrates a preference elicitation approach that allows this cost function formulation to capture a user’s subjective preferences regarding how his or her user interfaces should be generated. The second cost function (Section 5.2) reflects the expected time a person would need to perform atypical set of tasks with a particular user interface. This cost function can capture a person’s objective motor abilities [27] and allows user interfaces to be directly optimized for speed of use. The last part of this section describes an extension that enables a notion of presentation consistency to be included as one of the terms in the cost function. 5.1. Factorization fore cient computation and personalization To develop a cost function that supports fast performance of the optimization algorithm as well as personalization, we start with three design requirements. As discussed in Section 3.3, to enable generating user interfaces adapted to a person’s usage patterns, the cost function should take into account information from usage traces, so as to provide an estimate of the expected cost with respect to the actual or anticipated usage. This is an effective mechanism for allowing some parts of an interface to be considered more important than others without forcing the designer to embed such information in the functional specification itself. To enable thee cient computation of the admissible heuristic on which the optimization algorithm relies (Table line 2), we require that the cost function be factorable as a sum of costs overall elements in the functional specification. That way, the cost of already assigned variables can be computed exactly, and for the remaining variables, the cost of the best feasible widget (i.e., one with smallest cost and which has not been pruned by constraint propagation) is used. To support personalization, the cost function should be parametrized in such away that the appropriate choice of parameters can result indifferent styles of user interfaces being favored over others. K.Z. Gajos et al. / Artificial Intelligence 174 (2010) 910–950 921 Download 5.78 Mb. Share with your friends:
|