Measuring patent similarity with SAO semantic analysis

Abstract

Patents are not only an important aspect of intellectual property rights, but they are also one of the only ways to protect technological inventions. However, in recent years, the number of patents has been increasing dramatically and, as a result, both patent applicants and patent examiners are finding it more difficult to conduct the due diligence step of the patent registration process. Therefore, the lack of a quick and easy way to accurately measure patent similarity has become a significant obstacle to protecting intellectual property. Currently, there are three main ways to measure patent similarity: IPC code analysis, citation analysis, and keyword analysis. None of these approaches are able to fully reflect the semantics in a patent's content. As an emerging methodology, subject-action-object (SAO) semantic analysis does reflect semantics, but most approaches treat each identified relationship as equally important, which does not necessarily provide an accurate measure of patent similarity. To offer this power to SAO analysis, this article introduces a new indicator called DWSAO as a reflection of the weight of each SAO semantic structure. Further, we present a semantic analysis framework that incorporates the DWSAO index for finding similar patents based on the weight of each SAO structure in the patent. A case study on the similarity of patents in the field of robotics was used to verify the reliability of the method. The results highlight the detailed meanings derived from the method, the accuracy of the outcomes, and the practical significance of using this approach.

Generic SAO Similarity Measure via Extended Sørensen-Dice Index

“…SAO is a syntactic structure that expresses the semantic relationship between things, i.e., how the entity subject (S) of a sentence relates to the entity object (O) of a sentence through an entity action (A) [7]. Subjects can represent ''solutions'', actions can represent either the ''effect'' or the ''influence'' of the solution, and objects can represent the ''invention problem'' [49].…”

“…Using the SAO structures to exploit the technological content of patents has significant advantages over traditional patent features [7], [50]. Hence, there is an increasing interest in studying the SAO semantic similarity metric, which has been widely used for various patent analyses, e.g., patent infringement identification [28]- [31].…”

“…This is due to the inherent flexibility of natural language enabling to express similar meanings using quite different sentences in terms of structure and word content. Thus, the SAO semantic similarity is usually measured by the following acknowledged Sørensen-Dice index exploiting the information of the underlying semantic similarity among elements in sets [7]. Given two sets X = {x 1 , .…”

“…Semantic similarity analysis is an indispensable module for applications in natural language processing (NLP) and related areas [1], such as text mining [2], information retrieval [3], machine learning [4], [5], and patent analysis [6], [7]. The measure of semantic similarity can be defined as a metric assessing the degree to which two texts are similar to each other in terms of meaning.…”

“…However, as shown by Wang et al [7], the performance of this commonly used method is far from desirable due to the relatively high recurrence rate and poor discrimination. Such a situation is caused by the fact that the number of terms (words or phrases) in the SAO structures is small.…”