Automated classification of A/E/C web content

In this paper the adaptation of a standard information retrieval technique, namely latent semantic indexing, is examined for a domain specific search engine. The premise behind this approach is that it is possible to accurately identify classification codes related to the content of the web page or web site. If content can be accurately classified then a user searching for content in a particular area (e.g. by specifying a classification code) will be presented only with highly relevant web information. The reason that we attempt to classify to a standard classification code is that these are used and understood by the vast majority of professionals within the A/E/C industries. Because a classification code has a well described scope it is likely to be understood similarly by professionals from many disciplines. Therefore, a system that can accurately retrieve information associated with a classification code is one which can be tied to many processes within the A/E/C profession where information is associated with these codes. This paper describes the ongoing development of the LSI-based search engine. It concentrates particularly on the testing of the resultant search engine in terms of the precision of the classification of construction industry web pages to a construction industry classification system (by comparison with an expert’s determination of correct classifications). It also provides an analysis of the developed search engine’s search result accuracy by comparison with the results returned by other major search engines in current use (e.g. Google and Yahoo) on the same query formulation.