專利技術功效矩陣之建構─特徵值選取方法之研究

The Investigation of Feature Selection Methods on the Effectiveness of Patent Technology / Function Matrix Construction

林昱智(Yu-Chih Lin)

指導教授 ： 楊錦生

元智大學/資訊學院/資訊管理學系/碩士(2011年)

摘要

隨著全球經濟的蓬勃發展，各大企業間的競爭日益激烈，專利侵權事件更是層出不窮，因此近年來智慧財產權受到了各大企業的重視，其中專利(Patent)更是扮演了舉足輕重的角色，而將技術專利化可以保護企業的技術成果與商業利益，甚至透過有效的專利佈局，使企業保有一定競爭優勢。拜資訊科技的進步所賜，知識取得也更加容易，全球的專利數量每年不斷的急遽增加，要如何在眾多的專利之中，快速地將專利資訊變成有用的專利情報、勾勒出專利地圖(Patent Map)，並進一步利用其分析結果來做迴避設計 (Design Around)，以免誤侵別人設好的專利地雷(Patent Mine)陷阱，或以挖洞技術聚焦利基點來強化公司之專利佈局等，就成了重要的課題。專利地圖主要可分為：專利管理圖(Management Charts)、專利技術圖(Technique Charts)、專利引證圖(Citation Charts)、以及專利權利要件圖(Claim Charts)等。其中，專利技術圖可用來了解競爭者之相關技術領域及未來企業專利佈局方向等。專利工程師若要將大量的專利文件一件件歸類到適當的技術與功效類別，往往需耗費許多時間與人力。故本研究主要目的為利用文字探勘(Text Mining)技術，配合TF、TF-IDF、Chi-Square、Entropy、Mutual Information等特徵值選取方法將專利資料進行自動歸類，分析出各個專利適合的技術與功效類別，協助專利工程師於專利技術功效矩陣圖(Technology Function Matrix)之建構，有效降低專利工程師在整理專利資料所花費的時間，進而提升企業在相關技術領域中專利佈局的腳步。

關鍵字

技術功效矩陣圖 ； 文字探勘 ； 特徵值選取 ； 專利地圖 ； 專利分析

並列摘要

In the emerging knowledge economy era, competition between companies is diverse, dynamic, and intense. To gain sustainable competitiveness, companies pay great attention to the development, deployment, and management of their intelligent properties (especially, patents). For example, effective patent management can help the formulation of business strategies, such as patent portfolio and design around. A critical task in patent management is the construction of patent maps which offer tremendous business value by providing a visualized expression of inventions in a specific technological field. As the number of patents expands rapidly, the construction of patent maps becomes a time-consuming and labor-intensive task. As a result, this study concentrates on proposing a technique to facilitate the efficient and effective patent map construction. Among the various types of patent maps, we focus on the construction of technology function matrix because of its powerful capability of visualizing existing inventions in a specific technological field and the lack of research attentions on it. Specifically, a text mining approach is proposed to automatically analyze patent documents and then assign each patent into an appropriate technology (or function) category. Moreover, the effects of several state-of-the-art feature selection methods (i.e, TF, TF-IDF, Chi-Square, Entropy, and Mutual Information) on the proposed technology function matrix construction technique are also examined. The empirical evaluation results are promising and believed to be a beneficial support of patent engineers on constructing technology function matrix.

並列關鍵字

Patent Analysis ； Technology Function Matrix ； Text Mining ； Feature Selection ； Patent Map

參考文獻

7.冷耀世，專利實務論，全華圖書，2008年。

20.經濟部智慧財產局. http://www.tipo.gov.tw

15.巫啟台，文件之關聯資訊萃取及概念圖自動建構，國立成功大學資訊工程學系碩士論文，2002年。

17.曾致遠，淺談最大熵原理和統計物理學，物理雙月刊，第二十五卷，第三期，第467-475頁，2003。

21.Yoon, B.N., Yoon, C.B. and Park, Y.T., “On the Development and Application of a Self-organizing Feature Map-based Patent Map,” R&D Management, 32, 2002, pp.219-300.