Evaluating the effectiveness of keyword search strategy for patent identification

详细信息    查看全文

• 作者：Zhongquan Xie^a ; ^b ; ^{xiezhongquan@gmail.com} ; ^{xiezhongquan@126.com}Author Vitae ; Kumiko Miyazaki^a ; ^{miyazaki@mot.titech.ac.jp}Author Vitae
• 关键词：Patent identification ; Keywords search ; Effectiveness ; Patent claims ; Automotive software ; Embedded software
• 刊名：World Patent Information
• 出版年：2013
• 出版时间：March, 2013
• 年：2013
• 卷：35
• 期：1
• 页码：20-30
• 全文大小：357 K

文摘

The great usage of patent data in management of technology and innovation highlights the significance of patent identification. For identifying patents related to cross-disciplinary or integrated technologies or products that are emerging and cannot be clearly defined through patent classes or no definite related patent class exists in the patent system, keyword search is an appropriate identification method. However, literature has not explicitly addressed the effectiveness evaluation of keyword search strategy for patent identification, namely which parts of patent information (title, abstract, claims and description) should be used and how to select effective keywords to achieve a high level of effectiveness. This paper tries to fill this gap through evaluating the effectiveness of keyword search strategies of using different parts of patent information with different clusters of keywords, in the case of automotive software related patents issued by the USPTO. Clusters of keywords are selected through evaluating the effectiveness of each keyword by measuring recall and precision. Effectiveness of keyword search strategies of using different parts of patent information is evaluated with the indicators of type I error (missing patents that should be identified) and type II error (retrieving irrelevant patents). The results show that the most effective method of identifying patents in a specific domain through keyword search is using the patent information in the title, abstract and claims. This paper provides a principle of keywords selection for patent identification by using specific criteria of recall and precision.