Patentee Name Harmonisation - ecoom.be

DIAGNOSTICS
MITSUBISHIJUKOGYO MITSUBISHIKASEI 6 0,35 ≈
ALLERGANSALES ALLERGAN 5 0,38 ≈
CATERPILLAR CATERPILLARTRACTOR 7 0,39 ≈
KONINKLIJKEPHILIPSELECTRONIC PHILIPSELECTRONIC 11 0,39 =
CENTRENATIONALDELARECHERCHE
SCIENTIFIQUE
ETABLISSEMENTPUBLICDITCENTRE
NATIONALDELARECHERCHE
SCIENTIFIQUECNRS
26 0,40 =
If correct matches were to coincide with distinctive values in terms of the distance
measures, automated procedures could be envisaged. However, the results of this exercise do
not suggest such a pattern. An inspection of the distance values and validation reveals that
defining an unambiguous criterion value is not so straightforward. Moreover, one observes
immediately that several cases require a more in-depth analysis in order to define whether or
not both patentees are similar. Such an assessment immediately extends beyond name
harmonizing per se and is best categorized as an exercise in legal entity harmonization (see
section 2 - Patentee name harmonization and legal entity harmonization). As this approach is
clearly beyond the scope of the objectives envisaged in this contribution, it has not been
included in the final methodology. At the same time, it goes without saying that enriching
approximate string searching with address information seems highly promising when striving for
legal entity harmonization and might also contribute to automated name harmonizing. Both
extensions do imply however considerable validation efforts.
7 CONCLUSION
In this contribution, we have developed a comprehensive approach oriented towards
name harmonizing. Emphasis has been placed on maximizing the accuracy of procedures that
can be implemented automatically, i.e. without additional - and time-consuming - validation
efforts that require secondary information sources. Name variations are not combined if there is
any doubt that the names relate to different legal entities.
This has resulted in a transparent method whose outcome has been a reduced set of
harmonized names. At this stage, the total number of patentee names has been decreased by
17%. 13.5% of the harmonized names are matched with more than one original patentee
name, matching 2.6 names on average. When harmonizing takes place, the number of patents
allocated to the same entity increases on average by 7.2 patents in absolute terms (33.1% in
relative terms), signaling a considerable impact.
A detailed validation exercise was conducted for 35 harmonized names. Findings
revealed accuracy levels of 100% percent and a level of completeness of 99,62%. More details
about this validation can be found in Appendix 7.
EUROSTAT and its partners deliberately opted for a transparent method so that all
interested parties will be able to build further on the results obtained. In the belief that the
procedures described in this methodology can be further enriched and refined – and this also
applies to legal entity normalization - we would encourage activities in this direction.
Improving the accuracy levels for the methodology as a whole is feasible by introducing
expert assessments in a systematic manner. Given the volume of names involved, such an
effort is beyond the current resources of EUROSTAT and its partners who developed this
methodology (INCENTIM/SOOS, K.U.Leuven, and SOGETI). At the same time, numerous
researchers and analysts are currently working on name harmonizing efforts with specific
samples (e.g. technological fields, countries/regions, and sectors). For researchers engaged in
such efforts, building on this methodology might be helpful; equally, the insights obtained by
researchers and analysts might be beneficial for further refinement of the current methodology.
In other words, by sharing the methodology developed among the different communities
involved in patentee analysis, further improvements could be envisaged. Consequently,
EUROSTAT and its partners decided to put the complete methodology into the public domain.
In Appendix 1 and Appendix 2, the full set of procedures is made available, making
allowances for implementation, verification, and the development of appropriate
extensions. Furthermore, given its continuous involvement in the PATSTAT Taskforce activities,
16