The heuristic graph search was chosen for study and improvement, because a mostly working prototype was already implemented in VIPÊ . Thus it offered a basis for comparison with the other two examples. The approach of the prototype was also somewhat improved.
Simulated annealing (SA) was selected as an alternative approach to pagination. SA has been successful in other kinds of constraint satisfaction problems where a cost function must be minimized. SA was also tempting, because the heuristic approach seemed to have some serious limitations. Specifically the difficulty of guiding the search and designing and implementing new and better heuristics was considered difficult. In SA the goals may be coded as objectives in the cost function, and the balance between different goals is easily controllable.
Genetic algorithms (GA) were also considered. In literature discussing the differences between SA and GA it has been suggested that it would be wise to first try SA, and only if it doesn't give satisfactory results, the more complex GA should be applied. Thus, only SA was implemented in this work.
During the work, several different SA-implementations were attempted and discarded since they were too slow. Here two cases will be described, the first of which did not prove fast enough for practical use, while the second succeeded well. In addition to being experiments of pagination of editorial material, these cases demonstrate the difficulties of SA implementation, and hopefully offer insights on how to represent a problem successfully for SA.