Comparing case-based and backtrack search in a database application

Abstract

In this paper the relative performance of top-down, backtrack search is compared with bottom-up, case-based search. The methods used by these constraint satisfaction techniques are described with regard to the steps necessary to adapt these abstract techniques to a real-world domain. The tests are carried out on a large database drawn from the travel industry. The application is the problem of assembling a valid holiday package from the appropriate accommodation and transport inventory. For these techniques to be viable in a practical domain, they must also be seen to be reliable in that search times should not deviate excessively from the calculated average. The standard deviation of the complexity must be minimized as well as maximizing the average performance. There is good correlation between the distance metric as used by case-based search and the average and standard deviation of the search complexity. This information can be used to partition the problem domain around the distance corresponding to the mean for backtrack search.