Abstract
The development of modern computer science leads to information being generated speedily. More knowledge documents about patent are difficult to be classified consistently and promptly. In order to solve this problem, many specialists take document management, especially for technical reports and patent documents, as a significant research issue combining the expertise of IT, IP, and domain experts. In traditional categorization for patent documents, domain experts classify documents based on their experiences after reviewing documental contents. The development of automatic document categorization becomes an important research because traditional methods often bring inconsistent classification results. In this thesis, a categorization method using artificial neural network (ANN) is developed to classify patent documents based on pre-constructed ontology. Firstly, this system extracts the features of a document by using morphological analysis and sentence analysis. Secondly, these features are matched with classes and relations of pre-constructed ontology, and transferred into the inputs of ANN using two weight-transferring functions proposed in this research. Thirdly, a well-trained ANN model is applied to calculate and infer the relationships between given document and categories. We take two domains, Chemical Mechanical Polishing (CMP) and business knowledge documents, as our case studies to demonstrate the proposed system at work. International Patent Classification (IPC) is constructed as the classification schema and hierarchy.