download0 view144
twitter facebook

공공누리This item is licensed Korea Open Government License

dc.contributor.author
Andrew Rodriguez
dc.contributor.author
Myong K. Jeong
dc.contributor.author
고병열
dc.contributor.author
김병훈
dc.contributor.author
이재민
dc.date.accessioned
2019-08-28T07:41:41Z
dc.date.available
2019-08-28T07:41:41Z
dc.date.issued
2015-02-15
dc.identifier.issn
0957-4174
dc.identifier.uri
https://repository.kisti.re.kr/handle/10580/14395
dc.identifier.uri
http://www.ndsl.kr/ndsl/search/detail/article/articleSearchResultDetail.do?cn=NART70651044
dc.description.abstract
Identifying important patents helps to drive business growth and focus investment. In the past, centralitymeasures such as degree centrality and betweenness centrality have been applied to identify influentialor important patents in patent citation networks. How such a complex process like technological changecan be analyzed is an important research topic. However, no existing centrality measure leverages thepowerful graph kernels for this end. This paper presents a new centrality measure based on the changeof the node similarity matrix after leveraging graph kernels. The proposed approach provides a morerobust understanding of the identification of influential nodes, since it focuses on graph structure informationby considering direct and indirect patent citations. This study begins with the premise that thechange of similarity matrix that results from removing a given node indicates the importance of the nodewithin its network, since each node makes a contribution to the similarity matrix among nodes. Wecalculate the change of the similarity matrix norms for a given node after we calculate the singular valuesfor the case of the existence and the case of nonexistence of that node within the network. Then, the noderesulting in the largest change (i.e., decrease) in the similarity matrix norm is considered to be themost influential node. We compare the performance of our proposed approach with other widely-usedcentrality measures using artificial data and real-life U.S. patent data. Experimental results show thatour proposed approach performs better than existing methods.
dc.language
eng
dc.relation.ispartofseries
Expert Systems with Applications
dc.title
Graph kernel based measure for evaluation the influence of patents in a patent citation network
dc.citation.endPage
1486
dc.citation.number
3
dc.citation.startPage
1479
dc.citation.volume
42
dc.subject.keyword
Centrality measure
dc.subject.keyword
Patent citation network
dc.subject.keyword
Graph kernel
dc.subject.keyword
Similarity matrix
dc.subject.keyword
Matrix norm
Appears in Collections:
7. KISTI 연구성과 > 학술지 발표논문
Files in This Item:
There are no files associated with this item.

Browse