TY - GEN
T1 - Recommending missing citations for newly granted patents
AU - Oh, Sooyoung
AU - Lei, Zhen
AU - Lee, Wang-chien
AU - Yen, John
N1 - Publisher Copyright:
© 2014 IEEE.
PY - 2014/3/10
Y1 - 2014/3/10
N2 - The U.S. recently adopted a post-grant opposition procedure to encourage third parties to challenge the validity of newly granted patents by providing relevant prior patents that are missed during patent examination (i.e., missing citations). In this paper, we propose a recommendation system for missing citations for newly granted patents. The recommendation system, based on the patent citation network of a newly granted query patent, focuses on paths that start with the references of the query patent in the network. Our approach is to identify the relevancy of a candidate patent to the query patent by its citation relationship (paths) that are distinguished based on the direction, topology and semantics of the paths in the network. We consider six different types of paths between a candidate patent and a query patent based on their citation relationship and define a relevancy score for each path type. Accordingly, we rank candidate patents via a RankSVM model learned by using those relevancy scores as features. The experimental results show our approach significantly improves the average precision and recall performance compared to two baseline methods, i.e., Katz distance and text similarity.
AB - The U.S. recently adopted a post-grant opposition procedure to encourage third parties to challenge the validity of newly granted patents by providing relevant prior patents that are missed during patent examination (i.e., missing citations). In this paper, we propose a recommendation system for missing citations for newly granted patents. The recommendation system, based on the patent citation network of a newly granted query patent, focuses on paths that start with the references of the query patent in the network. Our approach is to identify the relevancy of a candidate patent to the query patent by its citation relationship (paths) that are distinguished based on the direction, topology and semantics of the paths in the network. We consider six different types of paths between a candidate patent and a query patent based on their citation relationship and define a relevancy score for each path type. Accordingly, we rank candidate patents via a RankSVM model learned by using those relevancy scores as features. The experimental results show our approach significantly improves the average precision and recall performance compared to two baseline methods, i.e., Katz distance and text similarity.
UR - http://www.scopus.com/inward/record.url?scp=84946691906&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=84946691906&partnerID=8YFLogxK
U2 - 10.1109/DSAA.2014.7058110
DO - 10.1109/DSAA.2014.7058110
M3 - Conference contribution
AN - SCOPUS:84946691906
T3 - DSAA 2014 - Proceedings of the 2014 IEEE International Conference on Data Science and Advanced Analytics
SP - 442
EP - 448
BT - DSAA 2014 - Proceedings of the 2014 IEEE International Conference on Data Science and Advanced Analytics
A2 - Karypis, George
A2 - Cao, Longbing
A2 - Wang, Wei
A2 - King, Irwin
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 2014 IEEE International Conference on Data Science and Advanced Analytics, DSAA 2014
Y2 - 30 October 2014 through 1 November 2014
ER -