Artificially Intelligent Specification and Analysis of Context-Dependent Attribute Preferences

4 Algorithms

If it is the case that the domain experts are dissatisfied with their present set of CDAP models, they may want some new models to adapt and work from. Typically these domain experts are fairly sure about the relevant properties in a particular market and the perceived characteristics of these properties for each product, what they are uncertain about is the number, identity and preferences of their customers. Thus there is a need for an algorithm which, given the relevant product characteristics, automatically searches for CDAP models that are consistent with the known data. An algorithm which we have found to be effective is described below - the Automatic CDAP Honing Engine (ACHE).

The heart of the procedure for determining a credible CDAP model from the sales data is a genetic programming algorithm. This is made more robust with a random search front-end to ensure a viable initial population of possible models and then a final hill-climbing evolutionary programming algorithm afterwards to tune the models found. This combination of three algorithms was found to produce qualitatively better solutions and work in a more robust manner than any of them separately. The robustness of this result is further enhanced by the implementation of the model in the logic-based programming language SDML (see [25]) which ensures that the individual algorithms and the integrated system are logically sound and consistent.

A binary-search hill-climbing algorithm optimizes the parameters of a single CDAP model, This algorithm can be used before and/or after ACHE at the discretion of the user.

4.1. - Genetic programming module
4.2. - Random Search
4.3. - Evolutionary programming

Artificially Intelligent Specification and Analysis of Context-Dependent Attribute Preferences - 03 NOV 97

[Next] [Previous] [Top] [Contents]

Generated with CERN WebMaker