美國專利文件為目前世界上涵蓋範圍最廣、類別數量最多的專利制度,但也因為其類別數量眾多且類別間又具有高度階層性質,導致相似專利文件搜尋不易與一般分類演算法無法呈現出專利文件類別所原有的階層關係,為了解決此一問題,本研究應用以增長層級式自我組織映射網路為基礎的分群架構搭配擷取SOM神經元權重值形成階層式文件分類器的方式,建構一相似專利文件搜尋系統,透過對使用者文件的歸類,縮小相似文件搜尋範圍並進行相似文件搜尋,提供相似專利文件予使用者參考。
The patent system of United States has the most extensive coverage and the largest number of categories in the world. Due to the large number of patent categories with highly hierarchy, it is hard to search similar patent document for a specified patent. Also, the hierarchical relationships among these patent documents are hardly expressed by using the general classification algorithms. In order to solve the difficulties, the GHSOM has been trained as a classifier to construct the patent documents suggestion system.