Abstract

Prior art retrieval plays an important role in patent examination. If a patent examiner can quickly and accurately retrieve prior art for an application patent, s/he will be able to efficiently and effectively judge the novelty of an application patent, and in turn, avoid hampering the technology development of the application domain. Moreover, in order to avoid losing tangible and intangible assets caused by patent infringement, companies need to search patent prior art before doing technology development. However, the increasing number of patents is not seen only in quantity but also in complexity. Bewaring the emerging need of prior art retrieval from large patent databases, the main objective of this study is to develop a methodology to effectively identify patent prior art. The developed prior art search framework firstly transforms claim sentences into the claim tree to represent the hierarchy of claimed elements, and then use the specification in patents to expand the terms in nodes of the claim tree. Finally, we calculate the similarity of claim trees and use the threshold to identify the most relevant patents as prior art. The prior art identified by the generated system was compared in recall, precision, and F-measure with those identified by the patent examiner. The results show the effectiveness of the proposed prior art retrieval system.

Share

COinS