Drivers of the decrease of patent similarities from 1976 to 2021.

Area of Science:

• Intellectual Property Analysis
• Bibliometrics
• Natural Language Processing (NLP)

Background:

• Patent citation networks reflect legal disclosure obligations.
• Textual similarity analysis is a method to study patent relationships.
• Existing patent similarity indicators show a declining trend since the mid-1970s.

Purpose of the Study:

• To investigate potential drivers of the observed decrease in patent textual similarity.
• To apply computationally efficient NLP tools for patent similarity scoring.
• To model patent similarity using generalized additive models.

Main Methods:

• Development of a computationally efficient patent similarity measure using state-of-the-art Natural Language Processing (NLP).
• Application of generalized additive models (GAMs) to model patent similarity scores.
• Analysis of temporally varying drivers influencing patent similarity levels.

Main Results:

• Non-linear modeling specifications successfully identified distinct, time-varying drivers of patent similarity.
• The proposed models explain a significant portion of the variation in similarity data (R² ≈ 18%), outperforming previous methods.
• The study reveals an underlying trend in patent similarity scores that differs from previously reported findings.

Conclusions:

• Advanced NLP and generalized additive models provide a more comprehensive understanding of patent similarity dynamics.
• The identified drivers and underlying trends offer a revised perspective on the evolution of patent textual similarity.
• This approach enhances the analysis of patent relationships and technological evolution.