Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Tracing the Flow of Knowledge From Science to Technology Using Deep Learning

Michael E. Rose, Mainak Ghosh, Sebastian Erhardt, Cheng Li, Erik Buunk, Dietmar Harhoff

TL;DR

The paper tackles the challenge of tracing knowledge flow from science to technology by building a cross-corpus transformer-based similarity framework that spans patents and scientific publications. It introduces Pat-SPECTER, a cross-corpus variant of SPECTER2Fine-tuned on patent data, and benchmarks it against PaECTER and other baselines using the Reliance on Science dataset (PPP and PPC) with ground-truth from patent office records. The authors demonstrate practical utility through validation tasks such as separating PPPs from PPCs and predicting the PPP paper among candidates, facilitated by the Logic Mill system and OpenAlex data. They also examine how duty of candor in different jurisdictions affects the semantic similarity of cited papers, offering a new quantitative lens on science-to-technology transfer and informing future knowledge-transfer analyses.

Abstract

We develop a language similarity model suitable for working with patents and scientific publications at the same time. In a horse race-style evaluation, we subject eight language (similarity) models to predict credible Patent-Paper Citations. We find that our Pat-SPECTER model performs best, which is the SPECTER2 model fine-tuned on patents. In two real-world scenarios (separating patent-paper-pairs and predicting patent-paper-pairs) we demonstrate the capabilities of the Pat-SPECTER. We finally test the hypothesis that US patents cite papers that are semantically less similar than in other large jurisdictions, which we posit is because of the duty of candor. The model is open for the academic community and practitioners alike.

Tracing the Flow of Knowledge From Science to Technology Using Deep Learning

TL;DR

The paper tackles the challenge of tracing knowledge flow from science to technology by building a cross-corpus transformer-based similarity framework that spans patents and scientific publications. It introduces Pat-SPECTER, a cross-corpus variant of SPECTER2Fine-tuned on patent data, and benchmarks it against PaECTER and other baselines using the Reliance on Science dataset (PPP and PPC) with ground-truth from patent office records. The authors demonstrate practical utility through validation tasks such as separating PPPs from PPCs and predicting the PPP paper among candidates, facilitated by the Logic Mill system and OpenAlex data. They also examine how duty of candor in different jurisdictions affects the semantic similarity of cited papers, offering a new quantitative lens on science-to-technology transfer and informing future knowledge-transfer analyses.

Abstract

We develop a language similarity model suitable for working with patents and scientific publications at the same time. In a horse race-style evaluation, we subject eight language (similarity) models to predict credible Patent-Paper Citations. We find that our Pat-SPECTER model performs best, which is the SPECTER2 model fine-tuned on patents. In two real-world scenarios (separating patent-paper-pairs and predicting patent-paper-pairs) we demonstrate the capabilities of the Pat-SPECTER. We finally test the hypothesis that US patents cite papers that are semantically less similar than in other large jurisdictions, which we posit is because of the duty of candor. The model is open for the academic community and practitioners alike.
Paper Structure (16 sections, 3 equations, 7 figures, 4 tables)

This paper contains 16 sections, 3 equations, 7 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Data
  3. Fine-tuning SPECTER2 and PaECTER
  4. Training Data
  5. Training
  6. Finding a cross-corpus language model
  7. Experimental setup
  8. Results
  9. Validation Exercises
  10. Separating Patent-Paper Pairs from Patent-Paper Citations
  11. Predicting Patent Paper Pairs in the patent-publication universe
  12. Data limitations
  13. From Science to Technology
  14. Interpretability
  15. Conclusion
  16. ...and 1 more sections

Figures (7)

  • Figure 1: Overview of the models used in the comparisons
  • Figure 2: KDE of the time lag between citing patent and cited paper
  • Figure 3: Distributions of Pat-SPECTER cosine similarities by type of pair
  • Figure 4: Empirical Cumulative Distribution of the rank for the matched PPPs
  • Figure 5: Distribution of the share of matched paper by patent
  • ...and 2 more figures