Knowledge Management through Content Interpretation

The improved performance of computer-based text analysis represents a major step forward for knowledge management. Reliable text interpretation allows focus to be placed upon the content of documents, rather than just the document wrapping, and this helps to emphasise the fundamental difference between knowledge management and document management. It is not uncommon for companies who wish to join the KM band-wagon to re-package existing document management programs with a KM label, even if such programs offer little more than a hierarchical file system and simple key-word search to support content management. In this paper we present CognIT's "Corporum" text analysis technology that is able to extract automatically the essential context of a given piece of text, and compare it with other texts to test whether they contain any overlap in contextual relevance. Therefore the technology can underpin several key knowledge management areas, including advanced search and retrieval, multidimensional text classification, meta-tagging, autosummarising, portal building, business intelligence, site surveillance etc. Performance is sufficient on a ordinary desktop PC to analyse 100s 1000s of texts per hour. The technology contains two essential elements. Firstly, the content of a given text is analysed thoroughly, and the contextual knowledge it contains is encapsulated automatically in the form of a detailed semantic net (ontology). The second element is then able to compare this knowledge representation with any other texts retrieved, using a neuro-fuzzy analysis method. This allows an estimate to be made of the document's relevance with respect to the original text, and also enables a justification of the analysis to be provided to the user in the form of a brief textual explanation outlining the relevance of the document. The current implementation of the technology is optimised for the English language. At present it is only able to interpret contextual relevance rather than intentionality. The system is not foolproof, but generally has a performance broadly comparable to a smart teenager. Ongoing research & development has already identified areas of major improvement.