Data sources

This study utilised several data sources (presented in Table 1). To retrieve information about technological developments, patent data was collected from the PATSTAT database, the Derwent⁷ Innovation Index (DII), the EuropePMC API and Relecura, which is a commercialised data source.

For retrieving and analysing the scientific literature, this study primarily relied on the Core Collection database provided by Thomson Reuters’ Web of Science (WOS).

PATSTAT contains raw patent data information extracted from patent documents stored in the EPO⁸’s master bibliographic database, called EPO worldwide bibliographic database (DOCDB). This database covers information from more than 90 patent authorities worldwide. Before 2016, PATSTAT used to be distributed on DVDs. The author can access PATSTA from the Lappeenranta University of Technology, as it has been installed locally. Retrieving patent data from the PATSTAT database requires some knowledge of database languages, such as SQL⁹. PATSTAT provides information about patent applications, publications, the names of applicants or inventors, citation information, patent families, technological categories and priority dates.

PATSTAT has several drawbacks. A significant issue is that the applicant and inventor names are not harmonised in PATSTAT (Kang and Tarasconi, 2016). This was a significant challenge for completing the tasks in Publication II, where the reviewers asked for detailed descriptive statistics of the inventors and companies involved in the dataset. Problematically, industrial or individual entities can appear in many formats or as different names. (Kang and Tarasconi, 2016) give the following example: Toyota Motor Corporation appears as “Toyota Motor Corporation,” “Toyota Motor Co,” “Toyota Jidosha Kabushiki Kaisha” (“Jidosha” and “Kabushiki Kaisha” mean a car and a corporation in Japanese, respectively), “Toyota Jidosha K. K.,” and so on. Similarly, “Yılmaz” and “Şahin,” which are Turkish common names, become “Yilmaz” or “Sahin” in the US patent office due to different character sets for documents written in English.

Entities with numerous appearances increase the difficulty for users trying to find and analyse all relevant patent data. To complete the task in Publication II, the list of companies was searched for manually by using and modifying google search engines. Another significant issue with patent databases involves the technological classification codes, as explained in section 2.4.2.

The Derwent Innovation Index (DII) is another source for patent data provided by Thompson Reuter’s database collection. It is accessible via the author’s university. This study used DII in Publication IV and Publication V. DII was used Publication IV because the data structure of patents is similar to that of scientific literature. The similarities between the tag fields and the data organisation made patent and publication data a comparable pair to create CPPAT to compare and visualise the scientific and patent landscapes. The unique structure of the patent abstracts created and maintained in

7 Worldwide Patent Statistical Database (PATSTAT).

8 European Patent Office.

9 Structured Query Language.

DII made this database useful for implementing the second phase of the analysis in Publication V.

Patent abstracts are re-written and divided to sub-sections in DII to facilitate prior art searches. The abstract subsections are detailed description, activity, use, advantage, drawing and novelty. The use of subsections highlights the section of the patent abstract relevant to possible applications or usage of the invention. In Publication V, the “use subsection” is retried from patent abstracts in DII and underwent text analytics process to reveal the technological application. However, a disadvantage of using DII or ISI is the limitations regarding downloading data on a large scale, as there is a limit of 500 records per download. Consequently, it takes time and effort for users to download the data and merge the records.

To conduct the quantitative analysis in Publication III, the authors used the EuropePMC¹⁰ database.

This database contains articles from the life sciences, as well as patents and clinical guidelines, which collectively provided a corpus of roughly 31.9 million records. Using a Python programming language script created by the authors, the database was accessed and text-mined using an Application Programming Interface (API). The API was used to search for documents with the terms “Taxol”

(Taxol’s commercial name) or “paclitaxel” (Taxol’s chemical name). The API was then used to retrieve the metadata foreach document, such as the authors, inventors, publication year and abstract.

One of the advantages of using EuropePMC API compared with DII or IST is that users can search, select and download both patent and publication data in a single query without any record limitations.

Relecura ¹¹ is a commercialised data source that was used in Publication V. The author’s co-authors at the Georgia Institute of Technology had access to Relecura. It was used to collect patent transaction data, which clarified the exchange of patent licences between industrial entities. Analysing the transaction data of sensor patents provided information regarding the industries and most traded/reassigned patents involved.

The Web of Science (WOS) Core Collection was the source for scientific publications utilised in this research. Relevant scientific publications on the quantitative study of STI were retrieved for the literature review in Publication I. One of WOS’ advantages compared to other databases is its coverage.

WOS contains records dating to 1900, making it a valuable data source for any prior art search.

Moreover, all journals and books covered by WOS are classified under a subject category created within the database. The WOS subject categories allow users to narrow down the results of their search query and remove all irrelevant documents. In Publication I, the search query was limited to the following subject categories (as practiced in previous literature reviews, i.e. (Martin et al., 2012):

business, management, economics, multidisciplinary science, social science interdisciplinary, operation research management science, information science, library science and computer science multidisciplinary application. This limitation allowed the authors to retrieve the documents relevant to the field of STI. The WOS was also used by scholars from the Georgia Institute of Technology in Publication V as a starting point for projecting the future industrial applications of triboelectric nanogenerators as a newly introduced technology (Fan, Tian, and Lin Wang, 2012).

10 https://europepmc.org/

11 https://relecura.com/

Table 1 Publications overview in terms of utilized data source and methods

15 https://www.thevantagepoint.com/

16 Natural Language Toolkit (NLTK) python library available at: https://www.nltk.org/

17 https://rapidminer.com/

Publication I Publication II Publication III Publication IV Publication V Data sources ₋ ISI Web of

Science (WOS)

₋ PATSTAT

PATSTAT EuropePMC API ₋ ISI Web of science

(WOS)

Expert opinion Expert opinion - Expert opinion

Case study ₋ Fuel cell electric vehicle

Taxol drug Online gaming Triboelectric

nanogenerators (TENG)

Analysis level

Technology level Technology level Technology level Technology level

Tools/

R libraries ₋ VantagePoint

₋ R libraries: TM, LDA, LDAvis