The matching exercise was performed between the Invest Europe data versus the EUIPO dataset, as well as between the Invest Europe data versus the PATSTAT dataset. The matching was based on two fields: the name and the country of origin of the applicant. The workflow applied to normalise the name of the applicant is described below:

• Remove empty or/and Not a Number (NaN) names.

• Transliterate (translate strings written in Greek or Cyrillic alphabet) the names.

• Remove symbols beyond simple letters or numbers.

• Translate accented to simple letters.

• Detect whether a legal form (based on the origin of country) exists and extract it.

The outcome of the above algorithm was the normalised name of the applicant, the legal form if it existed and its standardised form (e.g. for Netherlands, legal form “BV” and standardised form “Limited”).

Next, an absolute matching was executed based on the normalised applicant name and the country. As a final step, an approximate matching was performed using the two aforementioned fields and employing the Jaro-Winkler string metric (Jaro, 1989) as the similarity measure.