Abstract
In order to process huge amount of explicit knowledge documents in an organized manner, automatic document categorization is an important research area in the R&D knowledge management domain. In this paper, we propose a new document classification methodology based on neural network technology. We first extract key phrases from the document set by means of text processing and determine the significance of key phrases by their appearance frequency. After significant phrases are extracted, a keyword correlation analysis model is applied to compute similarity between key phrases. Then, synonyms are extracted from highly similar terms. The back-propagation neural network model is adopted as a classifier. The target output is to identify a document's proper category based on the hierarchical document classification scheme, i.e., the international patent classification (IPC) standard. In this research, patents related to designs of innovative power handtools are studied in their IPC classification scheme. Any related patent can be automatically and accurately classified using the pre-trained neural network models. In the prototype system, we provide two modules for explicit knowledge management. The automatic classification module helps the user classify patent documents and the search module helps users find the correct patent documents quickly. The result shows a very significant improvement in document classification and identification in explicit knowledge management.