Scholarlink

<p>This study aims at better mapping and explaining the linkages&#xa0;between science and technology, focusing on selected frontier technological domains (quantum cryptography, CRISPR, and CAR-T cells). How best to map science-technology linkages? Citation of a scientific publication by a patent (NPL: Non patent literature) and semantic similarity between the text of publication and a patent (NLP: Natural language processing) are two signals of possible knowledge transfer between science and technology. We compare these two signals for USPTO patents. We find only partial convergence between them when semantic proximity is calculated with the title and abstract of documents. We then calculate semantic proximity by using the full text of scientific publications and patent claims: this improves significantly the correspondence with citations. We find supportive evidence for two explanations for this discrepancy: (1) citations reflect commonality between specific ideas of documents, that do not necessarily show up in title and abstract, as shown by an analysis of full text of documents; (2) many citations in USPTO filings have weak relevance, they reflect informational limitations, legal constraints or strategic choices of the patent holder rather than knowledge transfer. The study brings two operational conclusions regarding the use of NLP for mapping science-technology linkages: (1) NLP on scientific publications and patents should use the full text of documents instead of the abstract; (2) we propose and test a new indicator of knowledge transfer, which combines NLP (on full text) and NPL and succeeds in exploiting their respective strengths.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

NPL v. NLP: analysing the links between science and technology with citations and semantics

  • 期刊论文

  • 发表日期：

    2025年11月6日

查看全文
  • Jianying Liu,
  • Mounir Amdaoud,
  • Wilfriedo Mescheba,
  • Justin Quemener,
  • David Sapinho,
  • Jean-Marc Deltorn,
  • Dominique Guellec

摘要

This study aims at better mapping and explaining the linkages between science and technology, focusing on selected frontier technological domains (quantum cryptography, CRISPR, and CAR-T cells). How best to map science-technology linkages? Citation of a scientific publication by a patent (NPL: Non patent literature) and semantic similarity between the text of publication and a patent (NLP: Natural language processing) are two signals of possible knowledge transfer between science and technology. We compare these two signals for USPTO patents. We find only partial convergence between them when semantic proximity is calculated with the title and abstract of documents. We then calculate semantic proximity by using the full text of scientific publications and patent claims: this improves significantly the correspondence with citations. We find supportive evidence for two explanations for this discrepancy: (1) citations reflect commonality between specific ideas of documents, that do not necessarily show up in title and abstract, as shown by an analysis of full text of documents; (2) many citations in USPTO filings have weak relevance, they reflect informational limitations, legal constraints or strategic choices of the patent holder rather than knowledge transfer. The study brings two operational conclusions regarding the use of NLP for mapping science-technology linkages: (1) NLP on scientific publications and patents should use the full text of documents instead of the abstract; (2) we propose and test a new indicator of knowledge transfer, which combines NLP (on full text) and NPL and succeeds in exploiting their respective strengths.