Automatically generating personalized user interfaces with Supple
Download 5.78 Mb. View original pdf
|
| Fig. 21. The classroom interface rendered fora small screen size (a) with an empty user trace (b) with a trace reflecting frequent transitions between individual light controls. Fig. 22. A fragment of the official Microsoft Ribbon (a) presented in a wide window (b) the same Ribbon fragment adapted to a narrower window some functionality is now contained in popup windows. terface generation problems, however, is highly discontinuous and therefore hard to investigate analytically. We, therefore, proceed with an empirical investigation. 7.4.1. Variable ordering heuristics and the parallel algorithm We empirically investigate both the average and the worst-case performance of Supple’s algorithm, using the factored version of the cost function described in Section 5. We start by investigating the properties of the three variable ordering heuristics considered in Section 4: bottom-up, top-down, and minimum remaining values (MRV). To examine a representative cross-section of the problem space for each interface considered, we pick two extreme screen size constraints one so large that a greedy approach to generating a user interface will succeed, the second just small enough that no interface can be generated for it. We interpolate at 100 intervals between these two extremes fora total of 101 screen sizes, and for each size we run the optimization algorithm, collecting the following measures: • The number of nodes expanded by the search algorithm before it finds the first solution and before it finds the best solution. • The time taken before the algorithm finds the first solution and before it finds the best solution. K.Z. Gajos et al. / Artificial Intelligence 174 (2010) 910–950 933 Download 5.78 Mb. Share with your friends:
|