Unsupervised learning based linkages between patents and scholarly publications

Arho Suominen, Hannes Toivanen

    Research output: Contribution to conferenceConference AbstractScientific


    Bibliometrics has been used to produce measures of knowledge flows between scholarly literature and patents, most notably by using the non-patent literature citation in patents. Existing methods offer a obstructed and narrow view of interplay between science and technology. This study complements existing methods by analyzing the semantic similarity of patents and publications in the context of Finland, uncovering thematic overlap between science and technology. The study uses Latent Dirichlet Allocations to analyze 185 931 patent and publication records in a merged corpus. The data spans patents (USPTO) and publications (WOS) with one or more Finnish author or inventor. The approach enabled the discovery of patent and publication links between documents without an explicit citation between the documents. This suggests that the method could complement existing approaches to science and technology mapping by producing a novel vantage point to the issue.
    Original languageEnglish
    Publication statusPublished - 2017
    EventAtlanta Conference on Science and Innovation Policy 2015 - Atlanta, United States
    Duration: 17 Sept 201519 Sept 2015
    Conference number: 6


    ConferenceAtlanta Conference on Science and Innovation Policy 2015
    Country/TerritoryUnited States


    Dive into the research topics of 'Unsupervised learning based linkages between patents and scholarly publications'. Together they form a unique fingerprint.

    Cite this