Building Patent Document Similarity by Considering the Dependency of Patent Classification Codes

• Main Authors: Kai-Yu Pu, 蒲開瑜
• Other Authors: 陳達仁
• Format: Others
• Language: en_US
• Published: 2010
• Online Access: http://ndltd.ncl.edu.tw/handle/45289681638383067792

碩士 === 臺灣大學 === 機械工程學研究所 === 98 === A methodology using patent classification codes as feature to calculate the patent document similarity which considered the feature dependency is presented in this study. The patent classification codes which is assigned by patent examiner as selected feature. Because each patent has several patent classification codes means they have feature dependency, and patent classification codes have their class schedule which is a hierarchical schedule. Two dependency factors will be introduced when we calculating the similarities of patent documents: One is the co-occurrence of classification codes which is the patent classification codes pair co-assigned in the same patent called co-occurrence; and another one is the hierarchical relation in patent classification codes schedule where each classification codes belong to is called hierarchical relation. However, some patent classification codes are infrequent so the co-occurrences are not all validity, so this study defines a threshold to filter the invalidity co-occurrences. The validity co-occurrences and hierarchical relations to show the dependencies of patent classification codes which belong to the selected patent data set are visualize by UCINET. Illustration of the systematic methodology is successfully demonstrated by using United State patent classification which is considered the hierarchical relation and co-occurrence as selected feature in some case studies.