Abstract
In an era of rapid information expansion, people encounter huge amount of intellectual property (IP), such as patents and knowledge documents in digital format. These documents are usually too many to be fully organized, read, understood and utilized. Therefore, efficient and effective ways of acquiring, organizing and presenting e-documents become very important for enterprises to manage their intellectual assets. In this thesis, a document (mainly patent document) summarization system using the integrated approach of key-phrase recognition and significant information density is proposed. First, a single document is uploaded, and key-phrases are recognize applying a sequence of text mining process such as stopping, stemming, term weight calculating, and phrase synthesizing. Second, based on similarity of paragraphs, paragraphs of the document are clustered to identify information concepts. Then, significant information density of each paragraph is measured. Significant information density is information masses that are calculated based on the summation of key-phrases, their relevant phrases, title phrases, domain-specific phrases, indicator phrases and topic sentences, divided by the total number of phrases in a paragraph or a document. Finally, key-phrases, paragraph that has highest density in each cluster, and metadata of document are picked into summary with well-designed summary template. External text mining game, compression ratio, retention ratio, and questionnaire survey are used in system experiment and evaluation. This research enables enterprises to organize knowledge and intellectual assets efficiently and to capture essence of knowledge documents effectively.