專利文件比對模型之研究

Research of Patent Matching System

謝周熾(Chou-Chih Hsiesh)

指導教授 ： 劉士豪

中原大學/商學院/資訊管理研究所/碩士(2005年)

https://doi.org/10.6840/cycu200500469

摘要

由於專利文件不僅僅提供企業或個人的智慧財產權在法律上的保護，更可作為一個企業在投入重大研發投資決策前，重要的決策依據，並藉由專利技術文獻縮短公司研發的時程，降低成本與風險等，進而提供企業長期競爭優勢。所以近似專利文件的檢索比對不論是在要行專利申請或是進行侵權比對分析時就顯得更為重要。過去不論是將探勘的技術應用於專利分類分析，或是將探勘分析應用於協助閱讀專利文件，都無法滿足能以專利比對出近似專利的需求。
專利文件本身具有的半結構化特性，專利在申請時就必須揭露的許多重要書目資訊，自然隱含許多可供比對參考的資訊。過去有許多研究將探勘分析技術應用到專利文件分類比對上，卻忽略掉了專利資訊本身有用的資訊，若能將專利裡面這些重要的資訊，應用到專利文件的比對上，或許可以提供一個更好的專利文件比對方法。
本研究提出一個專利比對的模型，除了將Text Mining應用在專利比對之外，同時將書目比對的模型架構在Text Mining比對的模型之上。觀察本研究的實驗結果，我們可以發現單以Text Mining對大量的產業專利進行分析，其分析結果已經有相當程度的比對效益；並且當預期比對的專利資料集合的相似程度越高時，Text Mining的比對效益亦跟著提高，不過當相似程度到達一定水準後，提高的程度就開始趨緩或不明顯。如果將比對模型加入書目比對進行分析後，我們可以發現專利比對的結果比單一使用Text Mining進行比對分析的方式有相當長足而明顯的改進。這證明了在本實驗的環境下，加入書目資訊比對，對於專利的比對分析有明顯的助益，也證明了本研究提出之加入書目比對方法的可行性。此外我們也可以發現到，加入書比對在專利資料集合的相似程度越高時，效益也有提升之現象。

關鍵字

書目比對 ； 文字探勘 ； 專利比對 ； 資料探勘

並列摘要

There were many researches about applying various data and text mining tools to patent analysis, and there were many scholars and experts having verified the accuracy and the feasibility of those data and text mining methodology in patent analysis. However, data and text mining methodology handled those patents as no more than “data” or “text”, and then tried to analyze the content using some methodology, maybe some linguistic algorithms, but they neglected some import features of patent itself. The Patent Matching System (PMS) is composing of patent’s bibliography matching engine and data mining matching engine. These two engines will generate a fusion of similarity rank table by weighting model, and then the related patents will be suggested to the user.
There are evidences that patent has become very important given by increasing lawsuits of patent. Accordingly, patent has become the critical weapon on the war of knowledge-based competition. If you have the “critical technology’s patent”, it means that you have the admission ticket to the victories.
Unfortunately, it was time-consuming for patent searchers to just find out the patents what he wanted; it was not only because the mass quantity of patent, but also needed the searcher’s specialty and experience to reduce the range of necessary patents by adopting various searching methods step by step. Even so, there were still many useless patents in the reduced collection of patents, and it was really a heavy job to read patents piece by piece.
There were many researches about applying various data and text mining tools to patent analysis, and there were many scholars and experts having verified the accuracy and the feasibility of those data and text mining methodology in patent analysis. However, data and text mining methodology handled those patents as no more than “data” or “text”, and then tried to analyze the content using some methodology, maybe some linguistic algorithms, but they neglected some import features of patent itself.
Because the feature of semi-structural content of patent data, it contains many critical information about each patent, such as citation’s information and patent assignees’ information etc. Trying to put more attention on that critical information and then trying to combine the feature of those patents with data and text mining tools may derive better efficiency.
Indeed, an aim of this article is trying to propose a patent matching architecture by combining the patent’s bibliography matching model with data and text mining matching model by weighting mode. After trying to build up a patent matching system (PMS) according to that architecture, we have verified the efficiency of the PMS.

並列關鍵字

data mining ； bibliography matching ； patent matching ； patent analysis

參考文獻

1. 組合自動化文件分類技術之研究-以專利文件分類為例–中原大學資管碩士論文-高志強-(2004)

2. 應用資料探勘分類技術於專利分析之研究–中原大學資管碩士論文-李駿祥 - (2003)

4. 李信穎，“專利地圖分析─電子商務軟體專利個案分析，” 中原大學資訊管理研究所碩士論文，2002。

24. 賴佳宏，「薄膜電晶體液晶顯示器(TFT-LCD)產業之技術發展趨勢研究—以專利分析與生命週期觀點—」，中原大學企業管理研究所碩士論文，2003。

1. Text mining applied to patent mapping: a practical business case Michele Fattori a,*, Giorgio Pedrazzi b, Roberta Turra b (2003)