This paper presents a hierarchical case representation that uses a context guided retrieval method. The performance of this method is compared to that of a simple flat file representation using standard nearest neighbour retrieval. The data presented in this paper is more extensive than that presented in an earlier paper by the same authors. The estimation of the construction costs of light industrial warehouse buildings is used as the test domain. Each case in the system comprises approximately 400 features. These are structured into a hierarchical case representation that holds more general contextual features at its top and specific building elements at its leaves. A modified nearest neighbour retrieval algorithm is used that is guided by contextual similarity. Problems are decomposed into sub-problems and solutions recomposed into a final solution. The comparative results show that the context guided retrieval method using the hierarchical case representation is significantly more accurate than the simpler flat file representation and standard nearest neighbour retrieval.
|Title of host publication||Research and development in expert systems XIV : Proceedings of Expert Systems 97, the seventeenth BCS SGES International Conference on Knowledge Based Systems and Applied Artificial Intelligence|
|Editors||John Hunt, Max Bramer|
|Place of Publication||United Kingdom|
|Number of pages||306|
|Publication status||Published - 1997|