KBS-class: A neural network tool for automatic content recognition of building texts

KBS-CLASS is a tool for automatic content recognition of building texts. It may be used for finding information in large or unstructured databases, for filtering news streams and as an aid for the classification of in-house-produced texts. There is likely to be a growing need for such tools as access to building knowledge in an electronic form becomes the key to an organization’s success. The KBS-CLASS tool is based on neural network technology. It is trained with texts which have been indexed according to a classification system. The tool is then able to give a contents' description of texts it has not yet seen, with terms adopted from the classification system. The tool should be able to use any classification system, such as SfB, R-UDC, BSAB or the new ISO system under development. The current tool has been trained with texts from a building products database from AB Svensk Byggtjänst (the Swedish Building Centre) using the BSAB system. A case study is presented to show how the tool may be used in a future work situation. The tool's performance is discussed. Finally, future directions for further development of the tool and similar tools are suggested.