Deep Learning for Tracing Knowledge Flow
Analysis
This paper introduces a novel language similarity model, Pat-SPECTER, for analyzing the relationship between scientific publications and patents. It's significant because it addresses the challenge of linking scientific advancements to technological applications, a crucial area for understanding innovation and technology transfer. The horse race evaluation and real-world scenario demonstrations provide strong evidence for the model's effectiveness. The investigation into jurisdictional differences in patent-paper citation patterns adds an interesting dimension to the research.
Key Takeaways
- •Developed Pat-SPECTER, a language similarity model for patents and scientific publications.
- •Demonstrated superior performance of Pat-SPECTER in predicting patent-paper citations.
- •Investigated jurisdictional differences in patent-paper citation patterns.
- •Model is open for academic and practical use.
Reference
“The Pat-SPECTER model performs best, which is the SPECTER2 model fine-tuned on patents.”